{
 "benchmark": "JevBench",
 "revision": "v1.2.8",
 "status": "final",
 "note": "Outcome codes: c correct, w wrong, f failed (scored wrong), n not attempted. Topic = item family. Per-task outcomes only for public items; held-out and imported items only in aggregates. Tier 'standard' and 'judge' accuracies are over all items of the tier (failures count as wrong), as in the scores.",
 "tasks": [
  {
   "id": "easy-intent-00",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-01",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-02",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-03",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-04",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-05",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-06",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-07",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-08",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-09",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-10",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-intent-11",
   "tier": "easy",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-fact-00",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-01",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-02",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-03",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-04",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-05",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-06",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-07",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-08",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-09",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-10",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-fact-11",
   "tier": "easy",
   "topic": "fact",
   "public": true,
   "type": "noul"
  },
  {
   "id": "easy-extraction-00",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-01",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-02",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-03",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-04",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-05",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-06",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-07",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-08",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-09",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-10",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-extraction-11",
   "tier": "easy",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-00",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-01",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-02",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-03",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-04",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-05",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-06",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-07",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-08",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-09",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-10",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "easy-tool_selection-11",
   "tier": "easy",
   "topic": "tool_selection",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-policy-01-0",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-01-1",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-02-0",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-02-1",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-03-0",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-03-1",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-04-0",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-04-1",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-05-0",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-05-1",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-06-0",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-policy-06-1",
   "tier": "standard",
   "topic": "policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-intent-01-0",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-01-1",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-02-0",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-02-1",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-03-0",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-03-1",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-04-0",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-04-1",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-05-0",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-05-1",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-06-0",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-intent-06-1",
   "tier": "standard",
   "topic": "intent",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-ordinal-01-0",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-01-1",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-02-0",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-02-1",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-03-0",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-03-1",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-04-0",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-04-1",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-05-0",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-05-1",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-06-0",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-ordinal-06-1",
   "tier": "standard",
   "topic": "ordinal",
   "public": true,
   "type": "score"
  },
  {
   "id": "original-extraction-01-0",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-01-1",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-02-0",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-02-1",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-03-0",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-03-1",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-04-0",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-04-1",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-05-0",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-05-1",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-06-0",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-extraction-06-1",
   "tier": "standard",
   "topic": "extraction",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-adequacy-01-0",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-01-1",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-02-0",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-02-1",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-03-0",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-03-1",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-04-0",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-04-1",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-05-0",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-05-1",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-06-0",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-adequacy-06-1",
   "tier": "standard",
   "topic": "adequacy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "original-routing-01-0",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-01-1",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-02-0",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-02-1",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-03-0",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-03-1",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-04-0",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-04-1",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-05-0",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-05-1",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-06-0",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "original-routing-06-1",
   "tier": "standard",
   "topic": "routing",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-long_policy-01",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-long_policy-04",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-long_policy-08",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-long_policy-09",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-long_policy-11",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-long_policy-13",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-long_policy-17",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-long_policy-19",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-probability-03",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-probability-04",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-probability-07",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-probability-08",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-temporal_numeric-03",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-temporal_numeric-06",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-temporal_numeric-07",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-a-temporal_numeric-09",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-a-temporal_numeric-12",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "score"
  },
  {
   "id": "hard-opus-b-ambiguous-02",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-ambiguous-03",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-ambiguous-07",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-ambiguous-09",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-ambiguous-10",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-ambiguous-11",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-ambiguous-13",
   "tier": "hard",
   "topic": "ambiguous",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-multi_hop-03",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-multi_hop-04",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-multi_hop-05",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "score"
  },
  {
   "id": "hard-opus-b-multi_hop-07",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-multi_hop-08",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-b-probability-01",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-probability-02",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-probability-03",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-b-probability-04",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-probability-06",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-tradeoff-01",
   "tier": "hard",
   "topic": "tradeoff",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-tradeoff-03",
   "tier": "hard",
   "topic": "tradeoff",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-tradeoff-06",
   "tier": "hard",
   "topic": "tradeoff",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-b-tradeoff-07",
   "tier": "hard",
   "topic": "tradeoff",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-b-tradeoff-08",
   "tier": "hard",
   "topic": "tradeoff",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-b-tradeoff-12",
   "tier": "hard",
   "topic": "tradeoff",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-long_policy-02",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-long_policy-03",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-long_policy-04",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-c-long_policy-05",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "score"
  },
  {
   "id": "hard-opus-c-long_policy-08",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "score"
  },
  {
   "id": "hard-opus-c-long_policy-10",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-long_policy-11",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-c-probability-03",
   "tier": "hard",
   "topic": "probability",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-temporal_numeric-02",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-opus-c-temporal_numeric-03",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-temporal_numeric-04",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-temporal_numeric-06",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-temporal_numeric-08",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-opus-c-temporal_numeric-12",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-adversarial-01",
   "tier": "hard",
   "topic": "adversarial",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-adversarial-06",
   "tier": "hard",
   "topic": "adversarial",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-a-adversarial-07",
   "tier": "hard",
   "topic": "adversarial",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-adversarial-08",
   "tier": "hard",
   "topic": "adversarial",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-a-adversarial-09",
   "tier": "hard",
   "topic": "adversarial",
   "public": true,
   "type": "score"
  },
  {
   "id": "hard-sol-a-adversarial-11",
   "tier": "hard",
   "topic": "adversarial",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-01",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-02",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-05",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-07",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-08",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-09",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-10",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-multi_hop-12",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-trap-02",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-a-trap-04",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-a-trap-05",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "score"
  },
  {
   "id": "hard-sol-a-trap-06",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-trap-08",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-trap-10",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-a-trap-13",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-a-trap-15",
   "tier": "hard",
   "topic": "trap",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-01",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-02",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-05",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-08",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-10",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-14",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-15",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-judge_hard-18",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-b-long_policy-01",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-long_policy-02",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-long_policy-05",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-long_policy-06",
   "tier": "hard",
   "topic": "long_policy",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-routing_hard-01",
   "tier": "hard",
   "topic": "routing_hard",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-routing_hard-02",
   "tier": "hard",
   "topic": "routing_hard",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-routing_hard-03",
   "tier": "hard",
   "topic": "routing_hard",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-routing_hard-07",
   "tier": "hard",
   "topic": "routing_hard",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-routing_hard-09",
   "tier": "hard",
   "topic": "routing_hard",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-temporal_numeric-01",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-temporal_numeric-02",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-temporal_numeric-03",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-b-temporal_numeric-04",
   "tier": "hard",
   "topic": "temporal_numeric",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-c-judge_hard-03",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-05",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-07",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-08",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-09",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-10",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-11",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-13",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-judge_hard-15",
   "tier": "hard",
   "topic": "judge_hard",
   "public": true,
   "type": "noul"
  },
  {
   "id": "hard-sol-c-multi_hop-07",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-c-multi_hop-09",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-c-multi_hop-11",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-c-multi_hop-12",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  },
  {
   "id": "hard-sol-c-multi_hop-13",
   "tier": "hard",
   "topic": "multi_hop",
   "public": true,
   "type": "choice"
  }
 ],
 "task_counts": {
  "easy": {
   "public": 48,
   "heldout_or_imported": 24
  },
  "standard": {
   "public": 72,
   "heldout_or_imported": 24
  },
  "judge": {
   "public": 0,
   "heldout_or_imported": 146
  },
  "hard": {
   "public": 111,
   "heldout_or_imported": 109
  }
 },
 "systems": {
  "semif-qwen3.5-4b": {
   "display": "SemIf, formerly OpenJev (Qwen3.5-4B, TheoLeeCJ)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 94,
     "w": 2,
     "f": 0,
     "n": 0,
     "accuracy": 0.9792
    },
    "judge": {
     "c": 139,
     "w": 7,
     "f": 0,
     "n": 0,
     "accuracy": 0.9521
    },
    "hard": {
     "c": 131,
     "w": 89,
     "f": 0,
     "n": 0,
     "accuracy": 0.5955
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 78,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 61,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8971
     }
    },
    "hard": {
     "long_policy": {
      "c": 16,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.4211
     },
     "probability": {
      "c": 8,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "temporal_numeric": {
      "c": 6,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     },
     "ambiguous": {
      "c": 8,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5714
     },
     "multi_hop": {
      "c": 22,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.6286
     },
     "tradeoff": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "adversarial": {
      "c": 10,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8333
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 26,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.7879
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.967
    ],
    "easy-intent-01": [
     "c",
     0.19
    ],
    "easy-intent-02": [
     "c",
     0.193
    ],
    "easy-intent-03": [
     "c",
     0.189
    ],
    "easy-intent-04": [
     "c",
     0.187
    ],
    "easy-intent-05": [
     "c",
     0.19
    ],
    "easy-intent-06": [
     "c",
     0.198
    ],
    "easy-intent-07": [
     "c",
     0.188
    ],
    "easy-intent-08": [
     "c",
     0.189
    ],
    "easy-intent-09": [
     "c",
     0.19
    ],
    "easy-intent-10": [
     "c",
     0.183
    ],
    "easy-intent-11": [
     "c",
     0.181
    ],
    "easy-fact-00": [
     "c",
     0.218
    ],
    "easy-fact-01": [
     "c",
     0.187
    ],
    "easy-fact-02": [
     "c",
     0.188
    ],
    "easy-fact-03": [
     "c",
     0.185
    ],
    "easy-fact-04": [
     "c",
     0.187
    ],
    "easy-fact-05": [
     "c",
     0.19
    ],
    "easy-fact-06": [
     "c",
     0.182
    ],
    "easy-fact-07": [
     "c",
     0.18
    ],
    "easy-fact-08": [
     "c",
     0.183
    ],
    "easy-fact-09": [
     "c",
     0.179
    ],
    "easy-fact-10": [
     "c",
     0.181
    ],
    "easy-fact-11": [
     "c",
     0.186
    ],
    "easy-extraction-00": [
     "c",
     0.181
    ],
    "easy-extraction-01": [
     "c",
     0.192
    ],
    "easy-extraction-02": [
     "c",
     0.181
    ],
    "easy-extraction-03": [
     "c",
     0.19
    ],
    "easy-extraction-04": [
     "c",
     0.191
    ],
    "easy-extraction-05": [
     "c",
     0.187
    ],
    "easy-extraction-06": [
     "c",
     0.182
    ],
    "easy-extraction-07": [
     "c",
     0.185
    ],
    "easy-extraction-08": [
     "c",
     0.19
    ],
    "easy-extraction-09": [
     "c",
     0.182
    ],
    "easy-extraction-10": [
     "c",
     0.185
    ],
    "easy-extraction-11": [
     "c",
     0.182
    ],
    "easy-tool_selection-00": [
     "c",
     0.201
    ],
    "easy-tool_selection-01": [
     "c",
     0.188
    ],
    "easy-tool_selection-02": [
     "c",
     0.205
    ],
    "easy-tool_selection-03": [
     "c",
     0.202
    ],
    "easy-tool_selection-04": [
     "c",
     0.188
    ],
    "easy-tool_selection-05": [
     "c",
     0.194
    ],
    "easy-tool_selection-06": [
     "c",
     0.19
    ],
    "easy-tool_selection-07": [
     "c",
     0.186
    ],
    "easy-tool_selection-08": [
     "c",
     0.185
    ],
    "easy-tool_selection-09": [
     "c",
     0.191
    ],
    "easy-tool_selection-10": [
     "c",
     0.186
    ],
    "easy-tool_selection-11": [
     "c",
     0.186
    ],
    "original-policy-01-0": [
     "c",
     0.194
    ],
    "original-policy-01-1": [
     "c",
     0.19
    ],
    "original-policy-02-0": [
     "c",
     0.195
    ],
    "original-policy-02-1": [
     "w",
     0.188
    ],
    "original-policy-03-0": [
     "c",
     0.187
    ],
    "original-policy-03-1": [
     "c",
     0.188
    ],
    "original-policy-04-0": [
     "c",
     0.185
    ],
    "original-policy-04-1": [
     "c",
     0.182
    ],
    "original-policy-05-0": [
     "c",
     0.192
    ],
    "original-policy-05-1": [
     "c",
     0.186
    ],
    "original-policy-06-0": [
     "c",
     0.184
    ],
    "original-policy-06-1": [
     "c",
     0.188
    ],
    "original-intent-01-0": [
     "c",
     0.189
    ],
    "original-intent-01-1": [
     "c",
     0.192
    ],
    "original-intent-02-0": [
     "c",
     0.185
    ],
    "original-intent-02-1": [
     "c",
     0.188
    ],
    "original-intent-03-0": [
     "c",
     0.191
    ],
    "original-intent-03-1": [
     "c",
     0.182
    ],
    "original-intent-04-0": [
     "c",
     0.182
    ],
    "original-intent-04-1": [
     "c",
     0.183
    ],
    "original-intent-05-0": [
     "c",
     0.183
    ],
    "original-intent-05-1": [
     "c",
     0.182
    ],
    "original-intent-06-0": [
     "c",
     0.19
    ],
    "original-intent-06-1": [
     "c",
     0.192
    ],
    "original-ordinal-01-0": [
     "c",
     0.188
    ],
    "original-ordinal-01-1": [
     "c",
     0.183
    ],
    "original-ordinal-02-0": [
     "c",
     0.195
    ],
    "original-ordinal-02-1": [
     "c",
     0.188
    ],
    "original-ordinal-03-0": [
     "c",
     0.184
    ],
    "original-ordinal-03-1": [
     "c",
     0.188
    ],
    "original-ordinal-04-0": [
     "c",
     0.188
    ],
    "original-ordinal-04-1": [
     "c",
     0.181
    ],
    "original-ordinal-05-0": [
     "c",
     0.187
    ],
    "original-ordinal-05-1": [
     "c",
     0.191
    ],
    "original-ordinal-06-0": [
     "c",
     0.191
    ],
    "original-ordinal-06-1": [
     "c",
     0.186
    ],
    "original-extraction-01-0": [
     "c",
     0.182
    ],
    "original-extraction-01-1": [
     "c",
     0.179
    ],
    "original-extraction-02-0": [
     "c",
     0.191
    ],
    "original-extraction-02-1": [
     "c",
     0.182
    ],
    "original-extraction-03-0": [
     "c",
     0.195
    ],
    "original-extraction-03-1": [
     "c",
     0.19
    ],
    "original-extraction-04-0": [
     "c",
     0.182
    ],
    "original-extraction-04-1": [
     "c",
     0.191
    ],
    "original-extraction-05-0": [
     "c",
     0.189
    ],
    "original-extraction-05-1": [
     "c",
     0.185
    ],
    "original-extraction-06-0": [
     "c",
     0.183
    ],
    "original-extraction-06-1": [
     "c",
     0.187
    ],
    "original-adequacy-01-0": [
     "c",
     0.19
    ],
    "original-adequacy-01-1": [
     "c",
     0.185
    ],
    "original-adequacy-02-0": [
     "c",
     0.183
    ],
    "original-adequacy-02-1": [
     "c",
     0.176
    ],
    "original-adequacy-03-0": [
     "c",
     0.181
    ],
    "original-adequacy-03-1": [
     "c",
     0.184
    ],
    "original-adequacy-04-0": [
     "c",
     0.188
    ],
    "original-adequacy-04-1": [
     "c",
     0.187
    ],
    "original-adequacy-05-0": [
     "c",
     0.184
    ],
    "original-adequacy-05-1": [
     "c",
     0.184
    ],
    "original-adequacy-06-0": [
     "c",
     0.186
    ],
    "original-adequacy-06-1": [
     "c",
     0.194
    ],
    "original-routing-01-0": [
     "c",
     0.191
    ],
    "original-routing-01-1": [
     "c",
     0.194
    ],
    "original-routing-02-0": [
     "c",
     0.198
    ],
    "original-routing-02-1": [
     "c",
     0.186
    ],
    "original-routing-03-0": [
     "c",
     0.193
    ],
    "original-routing-03-1": [
     "c",
     0.185
    ],
    "original-routing-04-0": [
     "c",
     0.193
    ],
    "original-routing-04-1": [
     "c",
     0.186
    ],
    "original-routing-05-0": [
     "c",
     0.193
    ],
    "original-routing-05-1": [
     "c",
     0.195
    ],
    "original-routing-06-0": [
     "c",
     0.191
    ],
    "original-routing-06-1": [
     "c",
     0.198
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     1.603
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.487
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.448
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.473
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.439
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.447
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.454
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.432
    ],
    "hard-opus-a-probability-03": [
     "w",
     0.23
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.268
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.271
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.207
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     0.214
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.209
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.292
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.212
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.221
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.218
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.213
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.222
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.227
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.212
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.215
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.223
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.535
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.458
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.507
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.487
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.428
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.368
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.216
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.268
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.22
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.258
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     0.249
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.234
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.239
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.235
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.222
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.218
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.7
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.695
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.645
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.647
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.65
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     0.703
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.654
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.231
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.258
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.251
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.246
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.266
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.273
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.253
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.198
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.199
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.205
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.193
    ],
    "hard-sol-a-adversarial-09": [
     "w",
     0.189
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.196
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.451
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.452
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.441
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.442
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     0.432
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.434
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.447
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.434
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.199
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.185
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.205
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.214
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.2
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.199
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.188
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.206
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.195
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.195
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.197
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.211
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.193
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.198
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.19
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.188
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     0.547
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.542
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.537
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.539
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.204
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.205
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.199
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.205
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.199
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.197
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.193
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.196
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.206
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.201
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     0.199
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.193
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.194
    ],
    "hard-sol-c-judge_hard-09": [
     "w",
     0.203
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.205
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.205
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.201
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.198
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.473
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.469
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.481
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.22
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.48
    ]
   }
  },
  "openjev-razorback16": {
   "display": "OpenJev (DiffusionGemma 26B-A4B NVFP4, razorback16)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 92,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9583
    },
    "judge": {
     "c": 133,
     "w": 13,
     "f": 0,
     "n": 0,
     "accuracy": 0.911
    },
    "hard": {
     "c": 144,
     "w": 76,
     "f": 0,
     "n": 0,
     "accuracy": 0.6545
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     }
    },
    "judge": {
     "routing": {
      "c": 70,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8974
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 15,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.3947
     },
     "probability": {
      "c": 11,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.55
     },
     "temporal_numeric": {
      "c": 6,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     },
     "ambiguous": {
      "c": 10,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.7143
     },
     "multi_hop": {
      "c": 28,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8
     },
     "tradeoff": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 29,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.8788
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.256
    ],
    "easy-intent-01": [
     "c",
     0.157
    ],
    "easy-intent-02": [
     "c",
     0.162
    ],
    "easy-intent-03": [
     "c",
     0.233
    ],
    "easy-intent-04": [
     "c",
     0.161
    ],
    "easy-intent-05": [
     "c",
     0.24
    ],
    "easy-intent-06": [
     "c",
     0.238
    ],
    "easy-intent-07": [
     "c",
     0.232
    ],
    "easy-intent-08": [
     "c",
     0.23
    ],
    "easy-intent-09": [
     "c",
     0.162
    ],
    "easy-intent-10": [
     "c",
     0.162
    ],
    "easy-intent-11": [
     "c",
     0.161
    ],
    "easy-fact-00": [
     "c",
     0.236
    ],
    "easy-fact-01": [
     "c",
     0.234
    ],
    "easy-fact-02": [
     "c",
     0.23
    ],
    "easy-fact-03": [
     "c",
     0.238
    ],
    "easy-fact-04": [
     "c",
     0.254
    ],
    "easy-fact-05": [
     "c",
     0.246
    ],
    "easy-fact-06": [
     "c",
     0.246
    ],
    "easy-fact-07": [
     "c",
     0.167
    ],
    "easy-fact-08": [
     "c",
     0.159
    ],
    "easy-fact-09": [
     "c",
     0.162
    ],
    "easy-fact-10": [
     "c",
     0.243
    ],
    "easy-fact-11": [
     "c",
     0.246
    ],
    "easy-extraction-00": [
     "c",
     0.309
    ],
    "easy-extraction-01": [
     "c",
     0.239
    ],
    "easy-extraction-02": [
     "c",
     0.242
    ],
    "easy-extraction-03": [
     "c",
     0.251
    ],
    "easy-extraction-04": [
     "c",
     0.166
    ],
    "easy-extraction-05": [
     "c",
     0.165
    ],
    "easy-extraction-06": [
     "c",
     0.164
    ],
    "easy-extraction-07": [
     "c",
     0.157
    ],
    "easy-extraction-08": [
     "c",
     0.244
    ],
    "easy-extraction-09": [
     "c",
     0.23
    ],
    "easy-extraction-10": [
     "c",
     0.235
    ],
    "easy-extraction-11": [
     "c",
     0.231
    ],
    "easy-tool_selection-00": [
     "c",
     0.173
    ],
    "easy-tool_selection-01": [
     "c",
     0.166
    ],
    "easy-tool_selection-02": [
     "c",
     0.246
    ],
    "easy-tool_selection-03": [
     "c",
     0.241
    ],
    "easy-tool_selection-04": [
     "c",
     0.239
    ],
    "easy-tool_selection-05": [
     "c",
     0.249
    ],
    "easy-tool_selection-06": [
     "c",
     0.24
    ],
    "easy-tool_selection-07": [
     "c",
     0.236
    ],
    "easy-tool_selection-08": [
     "c",
     0.163
    ],
    "easy-tool_selection-09": [
     "c",
     0.246
    ],
    "easy-tool_selection-10": [
     "c",
     0.159
    ],
    "easy-tool_selection-11": [
     "c",
     0.229
    ],
    "original-policy-01-0": [
     "c",
     0.171
    ],
    "original-policy-01-1": [
     "c",
     0.231
    ],
    "original-policy-02-0": [
     "c",
     0.228
    ],
    "original-policy-02-1": [
     "c",
     0.229
    ],
    "original-policy-03-0": [
     "c",
     0.227
    ],
    "original-policy-03-1": [
     "c",
     0.221
    ],
    "original-policy-04-0": [
     "c",
     0.228
    ],
    "original-policy-04-1": [
     "c",
     0.23
    ],
    "original-policy-05-0": [
     "c",
     0.235
    ],
    "original-policy-05-1": [
     "c",
     0.244
    ],
    "original-policy-06-0": [
     "c",
     0.16
    ],
    "original-policy-06-1": [
     "c",
     0.248
    ],
    "original-intent-01-0": [
     "c",
     0.17
    ],
    "original-intent-01-1": [
     "c",
     0.254
    ],
    "original-intent-02-0": [
     "c",
     0.244
    ],
    "original-intent-02-1": [
     "c",
     0.233
    ],
    "original-intent-03-0": [
     "c",
     0.239
    ],
    "original-intent-03-1": [
     "c",
     0.169
    ],
    "original-intent-04-0": [
     "c",
     0.256
    ],
    "original-intent-04-1": [
     "c",
     0.254
    ],
    "original-intent-05-0": [
     "c",
     0.158
    ],
    "original-intent-05-1": [
     "c",
     0.16
    ],
    "original-intent-06-0": [
     "c",
     0.159
    ],
    "original-intent-06-1": [
     "c",
     0.243
    ],
    "original-ordinal-01-0": [
     "c",
     0.238
    ],
    "original-ordinal-01-1": [
     "c",
     0.233
    ],
    "original-ordinal-02-0": [
     "c",
     0.25
    ],
    "original-ordinal-02-1": [
     "c",
     0.249
    ],
    "original-ordinal-03-0": [
     "c",
     0.241
    ],
    "original-ordinal-03-1": [
     "c",
     0.231
    ],
    "original-ordinal-04-0": [
     "c",
     0.24
    ],
    "original-ordinal-04-1": [
     "c",
     0.235
    ],
    "original-ordinal-05-0": [
     "c",
     0.239
    ],
    "original-ordinal-05-1": [
     "c",
     0.247
    ],
    "original-ordinal-06-0": [
     "c",
     0.248
    ],
    "original-ordinal-06-1": [
     "c",
     0.253
    ],
    "original-extraction-01-0": [
     "c",
     0.174
    ],
    "original-extraction-01-1": [
     "c",
     0.169
    ],
    "original-extraction-02-0": [
     "c",
     0.162
    ],
    "original-extraction-02-1": [
     "c",
     0.252
    ],
    "original-extraction-03-0": [
     "c",
     0.248
    ],
    "original-extraction-03-1": [
     "c",
     0.246
    ],
    "original-extraction-04-0": [
     "c",
     0.159
    ],
    "original-extraction-04-1": [
     "c",
     0.162
    ],
    "original-extraction-05-0": [
     "c",
     0.23
    ],
    "original-extraction-05-1": [
     "c",
     0.162
    ],
    "original-extraction-06-0": [
     "c",
     0.234
    ],
    "original-extraction-06-1": [
     "c",
     0.155
    ],
    "original-adequacy-01-0": [
     "c",
     0.236
    ],
    "original-adequacy-01-1": [
     "c",
     0.239
    ],
    "original-adequacy-02-0": [
     "c",
     0.237
    ],
    "original-adequacy-02-1": [
     "c",
     0.231
    ],
    "original-adequacy-03-0": [
     "c",
     0.233
    ],
    "original-adequacy-03-1": [
     "w",
     0.229
    ],
    "original-adequacy-04-0": [
     "c",
     0.239
    ],
    "original-adequacy-04-1": [
     "c",
     0.247
    ],
    "original-adequacy-05-0": [
     "c",
     0.242
    ],
    "original-adequacy-05-1": [
     "c",
     0.158
    ],
    "original-adequacy-06-0": [
     "c",
     0.234
    ],
    "original-adequacy-06-1": [
     "c",
     0.248
    ],
    "original-routing-01-0": [
     "c",
     0.175
    ],
    "original-routing-01-1": [
     "c",
     0.232
    ],
    "original-routing-02-0": [
     "c",
     0.235
    ],
    "original-routing-02-1": [
     "c",
     0.226
    ],
    "original-routing-03-0": [
     "c",
     0.248
    ],
    "original-routing-03-1": [
     "c",
     0.245
    ],
    "original-routing-04-0": [
     "c",
     0.256
    ],
    "original-routing-04-1": [
     "w",
     0.241
    ],
    "original-routing-05-0": [
     "c",
     0.161
    ],
    "original-routing-05-1": [
     "c",
     0.167
    ],
    "original-routing-06-0": [
     "c",
     0.245
    ],
    "original-routing-06-1": [
     "c",
     0.239
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.633
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.449
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.445
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.447
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.342
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.442
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.447
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.422
    ],
    "hard-opus-a-probability-03": [
     "w",
     0.265
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.321
    ],
    "hard-opus-a-probability-07": [
     "c",
     0.309
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.252
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.261
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.278
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     0.343
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.273
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.286
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.196
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.278
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.266
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.263
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.266
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.254
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.271
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     0.466
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     0.426
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.436
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.448
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.445
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.396
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.26
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.31
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.195
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.302
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.291
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.279
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.284
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.285
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.287
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.29
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.602
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.597
    ],
    "hard-opus-c-long_policy-04": [
     "w",
     0.595
    ],
    "hard-opus-c-long_policy-05": [
     "c",
     0.586
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.601
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.625
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     0.566
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.267
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.298
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.303
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.292
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.313
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.303
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.295
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.254
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.254
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.254
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.243
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.235
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.174
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.349
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.346
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.421
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.446
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     0.446
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.417
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.431
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.43
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.244
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.239
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.258
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.269
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.18
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.269
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.175
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.269
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.249
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.17
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.172
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.163
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.248
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.166
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.168
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.236
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.424
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.399
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     0.491
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.53
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.266
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.27
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.27
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.265
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.266
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.257
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.277
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.266
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.256
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.26
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.174
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.242
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.256
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.258
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.258
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.257
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.245
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.24
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.451
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.442
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.447
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.261
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.473
    ]
   }
  },
  "system-one-sg": {
   "display": "system-one (Qwen3-8B, Sean Goedecke)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 87,
     "w": 9,
     "f": 0,
     "n": 0,
     "accuracy": 0.9062
    },
    "judge": {
     "c": 134,
     "w": 12,
     "f": 0,
     "n": 0,
     "accuracy": 0.9178
    },
    "hard": {
     "c": 110,
     "w": 110,
     "f": 0,
     "n": 0,
     "accuracy": 0.5
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "intent": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "routing": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     }
    },
    "judge": {
     "routing": {
      "c": 72,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9231
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 12,
      "w": 26,
      "f": 0,
      "n": 0,
      "accuracy": 0.3158
     },
     "probability": {
      "c": 10,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "temporal_numeric": {
      "c": 8,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.2667
     },
     "ambiguous": {
      "c": 5,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.3571
     },
     "multi_hop": {
      "c": 17,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4857
     },
     "tradeoff": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 20,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.6061
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.219
    ],
    "easy-intent-01": [
     "c",
     0.157
    ],
    "easy-intent-02": [
     "c",
     0.156
    ],
    "easy-intent-03": [
     "c",
     0.152
    ],
    "easy-intent-04": [
     "c",
     0.152
    ],
    "easy-intent-05": [
     "c",
     0.158
    ],
    "easy-intent-06": [
     "c",
     0.152
    ],
    "easy-intent-07": [
     "c",
     0.154
    ],
    "easy-intent-08": [
     "c",
     0.155
    ],
    "easy-intent-09": [
     "c",
     0.155
    ],
    "easy-intent-10": [
     "c",
     0.158
    ],
    "easy-intent-11": [
     "c",
     0.151
    ],
    "easy-fact-00": [
     "c",
     0.158
    ],
    "easy-fact-01": [
     "c",
     0.157
    ],
    "easy-fact-02": [
     "c",
     0.157
    ],
    "easy-fact-03": [
     "c",
     0.153
    ],
    "easy-fact-04": [
     "c",
     0.157
    ],
    "easy-fact-05": [
     "c",
     0.155
    ],
    "easy-fact-06": [
     "c",
     0.158
    ],
    "easy-fact-07": [
     "c",
     0.157
    ],
    "easy-fact-08": [
     "c",
     0.156
    ],
    "easy-fact-09": [
     "c",
     0.156
    ],
    "easy-fact-10": [
     "c",
     0.158
    ],
    "easy-fact-11": [
     "c",
     0.156
    ],
    "easy-extraction-00": [
     "c",
     0.151
    ],
    "easy-extraction-01": [
     "c",
     0.152
    ],
    "easy-extraction-02": [
     "c",
     0.157
    ],
    "easy-extraction-03": [
     "c",
     0.152
    ],
    "easy-extraction-04": [
     "c",
     0.158
    ],
    "easy-extraction-05": [
     "c",
     0.152
    ],
    "easy-extraction-06": [
     "c",
     0.153
    ],
    "easy-extraction-07": [
     "c",
     0.157
    ],
    "easy-extraction-08": [
     "c",
     0.152
    ],
    "easy-extraction-09": [
     "c",
     0.16
    ],
    "easy-extraction-10": [
     "c",
     0.157
    ],
    "easy-extraction-11": [
     "c",
     0.158
    ],
    "easy-tool_selection-00": [
     "c",
     0.15
    ],
    "easy-tool_selection-01": [
     "c",
     0.157
    ],
    "easy-tool_selection-02": [
     "c",
     0.156
    ],
    "easy-tool_selection-03": [
     "c",
     0.158
    ],
    "easy-tool_selection-04": [
     "c",
     0.153
    ],
    "easy-tool_selection-05": [
     "c",
     0.159
    ],
    "easy-tool_selection-06": [
     "c",
     0.16
    ],
    "easy-tool_selection-07": [
     "c",
     0.15
    ],
    "easy-tool_selection-08": [
     "c",
     0.153
    ],
    "easy-tool_selection-09": [
     "c",
     0.158
    ],
    "easy-tool_selection-10": [
     "c",
     0.15
    ],
    "easy-tool_selection-11": [
     "c",
     0.152
    ],
    "original-policy-01-0": [
     "c",
     0.173
    ],
    "original-policy-01-1": [
     "c",
     0.162
    ],
    "original-policy-02-0": [
     "c",
     0.16
    ],
    "original-policy-02-1": [
     "w",
     0.158
    ],
    "original-policy-03-0": [
     "c",
     0.156
    ],
    "original-policy-03-1": [
     "w",
     0.161
    ],
    "original-policy-04-0": [
     "c",
     0.158
    ],
    "original-policy-04-1": [
     "c",
     0.161
    ],
    "original-policy-05-0": [
     "c",
     0.161
    ],
    "original-policy-05-1": [
     "c",
     0.159
    ],
    "original-policy-06-0": [
     "c",
     0.154
    ],
    "original-policy-06-1": [
     "c",
     0.156
    ],
    "original-intent-01-0": [
     "c",
     0.157
    ],
    "original-intent-01-1": [
     "c",
     0.156
    ],
    "original-intent-02-0": [
     "w",
     0.159
    ],
    "original-intent-02-1": [
     "c",
     0.157
    ],
    "original-intent-03-0": [
     "c",
     0.157
    ],
    "original-intent-03-1": [
     "c",
     0.156
    ],
    "original-intent-04-0": [
     "c",
     0.156
    ],
    "original-intent-04-1": [
     "c",
     0.158
    ],
    "original-intent-05-0": [
     "c",
     0.159
    ],
    "original-intent-05-1": [
     "c",
     0.157
    ],
    "original-intent-06-0": [
     "c",
     0.155
    ],
    "original-intent-06-1": [
     "c",
     0.158
    ],
    "original-ordinal-01-0": [
     "c",
     0.156
    ],
    "original-ordinal-01-1": [
     "c",
     0.157
    ],
    "original-ordinal-02-0": [
     "c",
     0.159
    ],
    "original-ordinal-02-1": [
     "c",
     0.161
    ],
    "original-ordinal-03-0": [
     "c",
     0.159
    ],
    "original-ordinal-03-1": [
     "c",
     0.159
    ],
    "original-ordinal-04-0": [
     "c",
     0.157
    ],
    "original-ordinal-04-1": [
     "c",
     0.157
    ],
    "original-ordinal-05-0": [
     "c",
     0.157
    ],
    "original-ordinal-05-1": [
     "c",
     0.159
    ],
    "original-ordinal-06-0": [
     "c",
     0.158
    ],
    "original-ordinal-06-1": [
     "c",
     0.149
    ],
    "original-extraction-01-0": [
     "c",
     0.156
    ],
    "original-extraction-01-1": [
     "c",
     0.156
    ],
    "original-extraction-02-0": [
     "c",
     0.154
    ],
    "original-extraction-02-1": [
     "c",
     0.158
    ],
    "original-extraction-03-0": [
     "c",
     0.151
    ],
    "original-extraction-03-1": [
     "c",
     0.153
    ],
    "original-extraction-04-0": [
     "c",
     0.157
    ],
    "original-extraction-04-1": [
     "c",
     0.156
    ],
    "original-extraction-05-0": [
     "c",
     0.151
    ],
    "original-extraction-05-1": [
     "c",
     0.159
    ],
    "original-extraction-06-0": [
     "c",
     0.157
    ],
    "original-extraction-06-1": [
     "c",
     0.15
    ],
    "original-adequacy-01-0": [
     "c",
     0.149
    ],
    "original-adequacy-01-1": [
     "c",
     0.149
    ],
    "original-adequacy-02-0": [
     "c",
     0.159
    ],
    "original-adequacy-02-1": [
     "c",
     0.156
    ],
    "original-adequacy-03-0": [
     "w",
     0.158
    ],
    "original-adequacy-03-1": [
     "w",
     0.161
    ],
    "original-adequacy-04-0": [
     "c",
     0.156
    ],
    "original-adequacy-04-1": [
     "c",
     0.164
    ],
    "original-adequacy-05-0": [
     "w",
     0.15
    ],
    "original-adequacy-05-1": [
     "w",
     0.158
    ],
    "original-adequacy-06-0": [
     "c",
     0.155
    ],
    "original-adequacy-06-1": [
     "c",
     0.157
    ],
    "original-routing-01-0": [
     "c",
     0.151
    ],
    "original-routing-01-1": [
     "c",
     0.154
    ],
    "original-routing-02-0": [
     "c",
     0.15
    ],
    "original-routing-02-1": [
     "c",
     0.158
    ],
    "original-routing-03-0": [
     "c",
     0.16
    ],
    "original-routing-03-1": [
     "c",
     0.158
    ],
    "original-routing-04-0": [
     "c",
     0.157
    ],
    "original-routing-04-1": [
     "w",
     0.166
    ],
    "original-routing-05-0": [
     "c",
     0.154
    ],
    "original-routing-05-1": [
     "c",
     0.153
    ],
    "original-routing-06-0": [
     "c",
     0.158
    ],
    "original-routing-06-1": [
     "c",
     0.16
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.748
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.476
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.437
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.48
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.438
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.437
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.437
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.434
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.182
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.268
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.29
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.183
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.194
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.199
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.285
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.2
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.209
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.2
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.188
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.189
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.207
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.186
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.193
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.205
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.548
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.44
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.477
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.479
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     0.433
    ],
    "hard-opus-b-probability-01": [
     "w",
     0.364
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.189
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.271
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.207
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.256
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.238
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.225
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.228
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.218
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.209
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.209
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.702
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     0.684
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.637
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.655
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.653
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     0.685
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.65
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.206
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.248
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     0.241
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.233
    ],
    "hard-opus-c-temporal_numeric-06": [
     "c",
     0.256
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.259
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.232
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.168
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.165
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     0.173
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.162
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.166
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.175
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.435
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.442
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.437
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.438
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.434
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     0.43
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.435
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.439
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.172
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.159
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.175
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.183
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.169
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.157
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.162
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.171
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.172
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.16
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.161
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.158
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.157
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.158
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.161
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.161
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.567
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     0.531
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.541
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.536
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.182
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.173
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.169
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.172
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.173
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.166
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.172
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.159
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.159
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.183
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.18
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.173
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.169
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.18
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.182
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.183
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.18
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.171
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     0.47
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.472
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.482
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.193
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.478
    ]
   }
  },
  "jev-1.13.0": {
   "display": "Jev 1.13.0 (TypeSafe AI)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 95,
     "w": 1,
     "f": 0,
     "n": 0,
     "accuracy": 0.9896
    },
    "judge": {
     "c": 138,
     "w": 8,
     "f": 0,
     "n": 0,
     "accuracy": 0.9452
    },
    "hard": {
     "c": 163,
     "w": 57,
     "f": 0,
     "n": 0,
     "accuracy": 0.7409
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 74,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.9487
     },
     "adequacy": {
      "c": 64,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.9412
     }
    },
    "hard": {
     "long_policy": {
      "c": 23,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.6053
     },
     "probability": {
      "c": 16,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.8
     },
     "temporal_numeric": {
      "c": 8,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.2667
     },
     "ambiguous": {
      "c": 11,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.7857
     },
     "multi_hop": {
      "c": 30,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.8571
     },
     "tradeoff": {
      "c": 11,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9167
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 26,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.7879
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.649
    ],
    "easy-intent-01": [
     "c",
     0.656
    ],
    "easy-intent-02": [
     "c",
     0.694
    ],
    "easy-intent-03": [
     "c",
     0.779
    ],
    "easy-intent-04": [
     "c",
     0.638
    ],
    "easy-intent-05": [
     "c",
     0.666
    ],
    "easy-intent-06": [
     "c",
     0.734
    ],
    "easy-intent-07": [
     "c",
     0.638
    ],
    "easy-intent-08": [
     "c",
     0.685
    ],
    "easy-intent-09": [
     "c",
     0.688
    ],
    "easy-intent-10": [
     "c",
     0.642
    ],
    "easy-intent-11": [
     "c",
     0.779
    ],
    "easy-fact-00": [
     "c",
     0.705
    ],
    "easy-fact-01": [
     "c",
     0.611
    ],
    "easy-fact-02": [
     "c",
     0.652
    ],
    "easy-fact-03": [
     "c",
     0.619
    ],
    "easy-fact-04": [
     "c",
     0.684
    ],
    "easy-fact-05": [
     "c",
     0.703
    ],
    "easy-fact-06": [
     "c",
     0.644
    ],
    "easy-fact-07": [
     "c",
     0.659
    ],
    "easy-fact-08": [
     "c",
     0.691
    ],
    "easy-fact-09": [
     "c",
     0.738
    ],
    "easy-fact-10": [
     "c",
     0.685
    ],
    "easy-fact-11": [
     "c",
     0.624
    ],
    "easy-extraction-00": [
     "c",
     0.65
    ],
    "easy-extraction-01": [
     "c",
     0.625
    ],
    "easy-extraction-02": [
     "c",
     0.671
    ],
    "easy-extraction-03": [
     "c",
     0.603
    ],
    "easy-extraction-04": [
     "c",
     0.623
    ],
    "easy-extraction-05": [
     "c",
     0.682
    ],
    "easy-extraction-06": [
     "c",
     0.671
    ],
    "easy-extraction-07": [
     "c",
     0.628
    ],
    "easy-extraction-08": [
     "c",
     0.679
    ],
    "easy-extraction-09": [
     "c",
     0.67
    ],
    "easy-extraction-10": [
     "c",
     0.67
    ],
    "easy-extraction-11": [
     "c",
     0.627
    ],
    "easy-tool_selection-00": [
     "c",
     0.722
    ],
    "easy-tool_selection-01": [
     "c",
     0.672
    ],
    "easy-tool_selection-02": [
     "c",
     0.663
    ],
    "easy-tool_selection-03": [
     "c",
     0.643
    ],
    "easy-tool_selection-04": [
     "c",
     0.673
    ],
    "easy-tool_selection-05": [
     "c",
     0.649
    ],
    "easy-tool_selection-06": [
     "c",
     0.659
    ],
    "easy-tool_selection-07": [
     "c",
     0.638
    ],
    "easy-tool_selection-08": [
     "c",
     0.623
    ],
    "easy-tool_selection-09": [
     "c",
     0.744
    ],
    "easy-tool_selection-10": [
     "c",
     0.604
    ],
    "easy-tool_selection-11": [
     "c",
     0.686
    ],
    "original-policy-01-0": [
     "c",
     0.637
    ],
    "original-policy-01-1": [
     "c",
     0.623
    ],
    "original-policy-02-0": [
     "c",
     0.714
    ],
    "original-policy-02-1": [
     "c",
     0.605
    ],
    "original-policy-03-0": [
     "c",
     0.605
    ],
    "original-policy-03-1": [
     "c",
     0.618
    ],
    "original-policy-04-0": [
     "c",
     0.637
    ],
    "original-policy-04-1": [
     "c",
     0.661
    ],
    "original-policy-05-0": [
     "c",
     0.637
    ],
    "original-policy-05-1": [
     "c",
     0.63
    ],
    "original-policy-06-0": [
     "c",
     0.804
    ],
    "original-policy-06-1": [
     "w",
     0.604
    ],
    "original-intent-01-0": [
     "c",
     0.667
    ],
    "original-intent-01-1": [
     "c",
     0.668
    ],
    "original-intent-02-0": [
     "c",
     0.665
    ],
    "original-intent-02-1": [
     "c",
     0.66
    ],
    "original-intent-03-0": [
     "c",
     0.678
    ],
    "original-intent-03-1": [
     "c",
     0.606
    ],
    "original-intent-04-0": [
     "c",
     0.642
    ],
    "original-intent-04-1": [
     "c",
     0.612
    ],
    "original-intent-05-0": [
     "c",
     0.71
    ],
    "original-intent-05-1": [
     "c",
     0.654
    ],
    "original-intent-06-0": [
     "c",
     0.638
    ],
    "original-intent-06-1": [
     "c",
     0.687
    ],
    "original-ordinal-01-0": [
     "c",
     0.64
    ],
    "original-ordinal-01-1": [
     "c",
     0.642
    ],
    "original-ordinal-02-0": [
     "c",
     0.741
    ],
    "original-ordinal-02-1": [
     "c",
     0.636
    ],
    "original-ordinal-03-0": [
     "c",
     0.589
    ],
    "original-ordinal-03-1": [
     "c",
     0.652
    ],
    "original-ordinal-04-0": [
     "c",
     0.676
    ],
    "original-ordinal-04-1": [
     "c",
     0.632
    ],
    "original-ordinal-05-0": [
     "c",
     0.714
    ],
    "original-ordinal-05-1": [
     "c",
     0.672
    ],
    "original-ordinal-06-0": [
     "c",
     0.692
    ],
    "original-ordinal-06-1": [
     "c",
     0.666
    ],
    "original-extraction-01-0": [
     "c",
     0.649
    ],
    "original-extraction-01-1": [
     "c",
     0.666
    ],
    "original-extraction-02-0": [
     "c",
     0.693
    ],
    "original-extraction-02-1": [
     "c",
     0.678
    ],
    "original-extraction-03-0": [
     "c",
     0.65
    ],
    "original-extraction-03-1": [
     "c",
     0.654
    ],
    "original-extraction-04-0": [
     "c",
     0.673
    ],
    "original-extraction-04-1": [
     "c",
     0.663
    ],
    "original-extraction-05-0": [
     "c",
     0.651
    ],
    "original-extraction-05-1": [
     "c",
     0.746
    ],
    "original-extraction-06-0": [
     "c",
     0.61
    ],
    "original-extraction-06-1": [
     "c",
     0.676
    ],
    "original-adequacy-01-0": [
     "c",
     0.673
    ],
    "original-adequacy-01-1": [
     "c",
     0.64
    ],
    "original-adequacy-02-0": [
     "c",
     0.67
    ],
    "original-adequacy-02-1": [
     "c",
     0.665
    ],
    "original-adequacy-03-0": [
     "c",
     0.698
    ],
    "original-adequacy-03-1": [
     "c",
     0.65
    ],
    "original-adequacy-04-0": [
     "c",
     0.632
    ],
    "original-adequacy-04-1": [
     "c",
     0.683
    ],
    "original-adequacy-05-0": [
     "c",
     0.673
    ],
    "original-adequacy-05-1": [
     "c",
     0.688
    ],
    "original-adequacy-06-0": [
     "c",
     0.591
    ],
    "original-adequacy-06-1": [
     "c",
     0.605
    ],
    "original-routing-01-0": [
     "c",
     0.611
    ],
    "original-routing-01-1": [
     "c",
     0.632
    ],
    "original-routing-02-0": [
     "c",
     0.625
    ],
    "original-routing-02-1": [
     "c",
     0.689
    ],
    "original-routing-03-0": [
     "c",
     0.603
    ],
    "original-routing-03-1": [
     "c",
     0.64
    ],
    "original-routing-04-0": [
     "c",
     0.618
    ],
    "original-routing-04-1": [
     "c",
     0.688
    ],
    "original-routing-05-0": [
     "c",
     0.679
    ],
    "original-routing-05-1": [
     "c",
     0.653
    ],
    "original-routing-06-0": [
     "c",
     0.616
    ],
    "original-routing-06-1": [
     "c",
     0.636
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.866
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.649
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.69
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.666
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.687
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.645
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.717
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.653
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.694
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.724
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.615
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.683
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.668
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.627
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.634
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.706
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.684
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.631
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.658
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.622
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.661
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.644
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.667
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.693
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.666
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.652
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.684
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.658
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.607
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.648
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.652
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.623
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.641
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.633
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.612
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.713
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.675
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.629
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.631
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.657
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.867
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.849
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.813
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.843
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.906
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.823
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.801
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.685
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.683
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.667
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.679
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.62
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.693
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.639
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.639
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.642
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.64
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.737
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.661
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.635
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.77
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.675
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.666
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.641
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     0.679
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.71
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.696
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.676
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.663
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.673
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.676
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.619
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.68
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.62
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.68
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.692
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.713
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.711
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.709
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.679
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.728
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.749
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.659
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.684
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.866
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.818
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.823
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.939
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.663
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.628
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.631
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.699
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.624
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.635
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.621
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.705
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.68
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.672
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.643
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.644
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.681
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.672
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.704
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.659
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.631
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.694
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.641
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.68
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.655
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.659
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.604
    ]
   }
  },
  "system-one-open": {
   "display": "system-one-open (Gemma 4 E2B LoRA on an L4)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 90,
     "w": 6,
     "f": 0,
     "n": 0,
     "accuracy": 0.9375
    },
    "judge": {
     "c": 128,
     "w": 18,
     "f": 0,
     "n": 0,
     "accuracy": 0.8767
    },
    "hard": {
     "c": 108,
     "w": 112,
     "f": 0,
     "n": 0,
     "accuracy": 0.4909
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "intent": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "adequacy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "routing": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     }
    },
    "judge": {
     "routing": {
      "c": 70,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8974
     },
     "adequacy": {
      "c": 58,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.8529
     }
    },
    "hard": {
     "long_policy": {
      "c": 10,
      "w": 28,
      "f": 0,
      "n": 0,
      "accuracy": 0.2632
     },
     "probability": {
      "c": 9,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.45
     },
     "temporal_numeric": {
      "c": 8,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.2667
     },
     "ambiguous": {
      "c": 6,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.4286
     },
     "multi_hop": {
      "c": 15,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.4286
     },
     "tradeoff": {
      "c": 3,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "adversarial": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "trap": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "judge_hard": {
      "c": 25,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.7576
     },
     "routing_hard": {
      "c": 9,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     29.642
    ],
    "easy-intent-01": [
     "c",
     0.56
    ],
    "easy-intent-02": [
     "c",
     0.589
    ],
    "easy-intent-03": [
     "c",
     0.529
    ],
    "easy-intent-04": [
     "c",
     0.572
    ],
    "easy-intent-05": [
     "c",
     0.596
    ],
    "easy-intent-06": [
     "c",
     0.54
    ],
    "easy-intent-07": [
     "c",
     0.514
    ],
    "easy-intent-08": [
     "c",
     0.547
    ],
    "easy-intent-09": [
     "c",
     0.517
    ],
    "easy-intent-10": [
     "c",
     0.499
    ],
    "easy-intent-11": [
     "c",
     0.598
    ],
    "easy-fact-00": [
     "c",
     0.529
    ],
    "easy-fact-01": [
     "c",
     0.51
    ],
    "easy-fact-02": [
     "c",
     0.502
    ],
    "easy-fact-03": [
     "c",
     0.617
    ],
    "easy-fact-04": [
     "c",
     0.629
    ],
    "easy-fact-05": [
     "c",
     0.495
    ],
    "easy-fact-06": [
     "c",
     0.506
    ],
    "easy-fact-07": [
     "c",
     0.504
    ],
    "easy-fact-08": [
     "c",
     0.595
    ],
    "easy-fact-09": [
     "c",
     0.515
    ],
    "easy-fact-10": [
     "c",
     0.491
    ],
    "easy-fact-11": [
     "c",
     0.608
    ],
    "easy-extraction-00": [
     "c",
     0.495
    ],
    "easy-extraction-01": [
     "c",
     0.528
    ],
    "easy-extraction-02": [
     "c",
     0.508
    ],
    "easy-extraction-03": [
     "c",
     0.509
    ],
    "easy-extraction-04": [
     "c",
     0.537
    ],
    "easy-extraction-05": [
     "c",
     0.54
    ],
    "easy-extraction-06": [
     "c",
     0.512
    ],
    "easy-extraction-07": [
     "c",
     0.505
    ],
    "easy-extraction-08": [
     "c",
     0.509
    ],
    "easy-extraction-09": [
     "c",
     0.503
    ],
    "easy-extraction-10": [
     "c",
     0.5
    ],
    "easy-extraction-11": [
     "c",
     0.499
    ],
    "easy-tool_selection-00": [
     "c",
     0.508
    ],
    "easy-tool_selection-01": [
     "c",
     0.619
    ],
    "easy-tool_selection-02": [
     "c",
     0.594
    ],
    "easy-tool_selection-03": [
     "c",
     0.495
    ],
    "easy-tool_selection-04": [
     "c",
     0.534
    ],
    "easy-tool_selection-05": [
     "c",
     0.506
    ],
    "easy-tool_selection-06": [
     "c",
     0.521
    ],
    "easy-tool_selection-07": [
     "c",
     0.531
    ],
    "easy-tool_selection-08": [
     "c",
     0.51
    ],
    "easy-tool_selection-09": [
     "c",
     0.511
    ],
    "easy-tool_selection-10": [
     "c",
     0.501
    ],
    "easy-tool_selection-11": [
     "c",
     0.521
    ],
    "original-policy-01-0": [
     "c",
     29.405
    ],
    "original-policy-01-1": [
     "c",
     0.689
    ],
    "original-policy-02-0": [
     "c",
     0.708
    ],
    "original-policy-02-1": [
     "c",
     0.649
    ],
    "original-policy-03-0": [
     "w",
     0.643
    ],
    "original-policy-03-1": [
     "c",
     0.672
    ],
    "original-policy-04-0": [
     "w",
     0.691
    ],
    "original-policy-04-1": [
     "c",
     0.698
    ],
    "original-policy-05-0": [
     "c",
     0.714
    ],
    "original-policy-05-1": [
     "c",
     0.756
    ],
    "original-policy-06-0": [
     "c",
     0.636
    ],
    "original-policy-06-1": [
     "c",
     0.648
    ],
    "original-intent-01-0": [
     "w",
     0.747
    ],
    "original-intent-01-1": [
     "c",
     0.652
    ],
    "original-intent-02-0": [
     "c",
     0.641
    ],
    "original-intent-02-1": [
     "c",
     0.652
    ],
    "original-intent-03-0": [
     "c",
     0.645
    ],
    "original-intent-03-1": [
     "c",
     0.7
    ],
    "original-intent-04-0": [
     "c",
     0.663
    ],
    "original-intent-04-1": [
     "c",
     0.635
    ],
    "original-intent-05-0": [
     "c",
     0.644
    ],
    "original-intent-05-1": [
     "c",
     0.651
    ],
    "original-intent-06-0": [
     "c",
     0.726
    ],
    "original-intent-06-1": [
     "c",
     0.626
    ],
    "original-ordinal-01-0": [
     "c",
     0.649
    ],
    "original-ordinal-01-1": [
     "c",
     0.748
    ],
    "original-ordinal-02-0": [
     "c",
     0.65
    ],
    "original-ordinal-02-1": [
     "c",
     0.663
    ],
    "original-ordinal-03-0": [
     "c",
     0.645
    ],
    "original-ordinal-03-1": [
     "c",
     0.63
    ],
    "original-ordinal-04-0": [
     "c",
     0.654
    ],
    "original-ordinal-04-1": [
     "c",
     0.632
    ],
    "original-ordinal-05-0": [
     "c",
     0.679
    ],
    "original-ordinal-05-1": [
     "c",
     0.643
    ],
    "original-ordinal-06-0": [
     "c",
     0.659
    ],
    "original-ordinal-06-1": [
     "c",
     0.638
    ],
    "original-extraction-01-0": [
     "c",
     0.645
    ],
    "original-extraction-01-1": [
     "c",
     0.632
    ],
    "original-extraction-02-0": [
     "c",
     0.655
    ],
    "original-extraction-02-1": [
     "c",
     0.624
    ],
    "original-extraction-03-0": [
     "w",
     0.635
    ],
    "original-extraction-03-1": [
     "c",
     0.645
    ],
    "original-extraction-04-0": [
     "c",
     0.651
    ],
    "original-extraction-04-1": [
     "c",
     0.632
    ],
    "original-extraction-05-0": [
     "c",
     0.628
    ],
    "original-extraction-05-1": [
     "c",
     0.649
    ],
    "original-extraction-06-0": [
     "c",
     0.641
    ],
    "original-extraction-06-1": [
     "c",
     0.623
    ],
    "original-adequacy-01-0": [
     "c",
     0.628
    ],
    "original-adequacy-01-1": [
     "c",
     0.737
    ],
    "original-adequacy-02-0": [
     "c",
     0.644
    ],
    "original-adequacy-02-1": [
     "c",
     0.631
    ],
    "original-adequacy-03-0": [
     "c",
     0.636
    ],
    "original-adequacy-03-1": [
     "c",
     0.637
    ],
    "original-adequacy-04-0": [
     "c",
     0.634
    ],
    "original-adequacy-04-1": [
     "c",
     0.66
    ],
    "original-adequacy-05-0": [
     "c",
     0.657
    ],
    "original-adequacy-05-1": [
     "c",
     0.639
    ],
    "original-adequacy-06-0": [
     "c",
     0.712
    ],
    "original-adequacy-06-1": [
     "c",
     0.649
    ],
    "original-routing-01-0": [
     "c",
     0.664
    ],
    "original-routing-01-1": [
     "c",
     0.755
    ],
    "original-routing-02-0": [
     "c",
     0.638
    ],
    "original-routing-02-1": [
     "c",
     0.662
    ],
    "original-routing-03-0": [
     "c",
     0.624
    ],
    "original-routing-03-1": [
     "c",
     0.638
    ],
    "original-routing-04-0": [
     "c",
     0.656
    ],
    "original-routing-04-1": [
     "c",
     0.634
    ],
    "original-routing-05-0": [
     "c",
     0.643
    ],
    "original-routing-05-1": [
     "w",
     0.629
    ],
    "original-routing-06-0": [
     "c",
     0.645
    ],
    "original-routing-06-1": [
     "c",
     0.675
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     27.754
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.994
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     1.036
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.938
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.917
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.892
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.902
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     1.092
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.66
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.741
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.884
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.662
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.661
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.645
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.753
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.647
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.654
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.652
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.792
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.65
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.721
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.679
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.73
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.649
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     1.023
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.888
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.944
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     1.028
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     0.9
    ],
    "hard-opus-b-probability-01": [
     "w",
     0.832
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.648
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.705
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.688
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.686
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.685
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.677
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.664
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.674
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.66
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     0.658
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     1.147
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     1.116
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.169
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.103
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.091
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     1.168
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     1.223
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.657
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.682
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     0.663
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.677
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.689
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.675
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.657
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.639
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.633
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.655
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.637
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.637
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.729
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.975
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     1.054
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     0.948
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.989
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.942
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.961
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.961
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.938
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.663
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.711
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.645
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.641
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.637
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.64
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.67
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.637
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.73
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.642
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.641
    ],
    "hard-sol-b-judge_hard-08": [
     "c",
     0.648
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.637
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.634
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.724
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.628
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.977
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     0.96
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.937
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.944
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.721
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     0.654
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.692
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.66
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.638
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.653
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.635
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.638
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.651
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.628
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.649
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.75
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.642
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.648
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.651
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.647
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.635
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.67
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     0.95
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.934
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     0.963
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.64
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.041
    ]
   }
  },
  "nimble-9b": {
   "display": "Bespoke Nimble 9B (Bespoke Labs)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 91,
     "w": 5,
     "f": 0,
     "n": 0,
     "accuracy": 0.9479
    },
    "judge": {
     "c": 130,
     "w": 16,
     "f": 0,
     "n": 0,
     "accuracy": 0.8904
    },
    "hard": {
     "c": 144,
     "w": 76,
     "f": 0,
     "n": 0,
     "accuracy": 0.6545
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 67,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.859
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 23,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.6053
     },
     "probability": {
      "c": 12,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.6
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 7,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "multi_hop": {
      "c": 28,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8
     },
     "tradeoff": {
      "c": 7,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.5833
     },
     "adversarial": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 25,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.7576
     },
     "routing_hard": {
      "c": 9,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.342
    ],
    "easy-intent-01": [
     "c",
     0.344
    ],
    "easy-intent-02": [
     "c",
     0.346
    ],
    "easy-intent-03": [
     "c",
     0.374
    ],
    "easy-intent-04": [
     "c",
     0.372
    ],
    "easy-intent-05": [
     "c",
     0.371
    ],
    "easy-intent-06": [
     "c",
     0.385
    ],
    "easy-intent-07": [
     "c",
     0.376
    ],
    "easy-intent-08": [
     "c",
     0.38
    ],
    "easy-intent-09": [
     "c",
     0.376
    ],
    "easy-intent-10": [
     "c",
     0.382
    ],
    "easy-intent-11": [
     "c",
     0.384
    ],
    "easy-fact-00": [
     "c",
     0.371
    ],
    "easy-fact-01": [
     "c",
     0.368
    ],
    "easy-fact-02": [
     "c",
     0.363
    ],
    "easy-fact-03": [
     "c",
     0.365
    ],
    "easy-fact-04": [
     "c",
     0.363
    ],
    "easy-fact-05": [
     "c",
     0.354
    ],
    "easy-fact-06": [
     "c",
     0.354
    ],
    "easy-fact-07": [
     "c",
     0.356
    ],
    "easy-fact-08": [
     "c",
     0.347
    ],
    "easy-fact-09": [
     "c",
     0.361
    ],
    "easy-fact-10": [
     "c",
     0.356
    ],
    "easy-fact-11": [
     "c",
     0.356
    ],
    "easy-extraction-00": [
     "c",
     0.356
    ],
    "easy-extraction-01": [
     "c",
     0.353
    ],
    "easy-extraction-02": [
     "c",
     0.354
    ],
    "easy-extraction-03": [
     "c",
     0.349
    ],
    "easy-extraction-04": [
     "c",
     0.359
    ],
    "easy-extraction-05": [
     "c",
     0.361
    ],
    "easy-extraction-06": [
     "c",
     0.36
    ],
    "easy-extraction-07": [
     "c",
     0.358
    ],
    "easy-extraction-08": [
     "c",
     0.366
    ],
    "easy-extraction-09": [
     "c",
     0.366
    ],
    "easy-extraction-10": [
     "c",
     0.364
    ],
    "easy-extraction-11": [
     "c",
     0.354
    ],
    "easy-tool_selection-00": [
     "c",
     0.376
    ],
    "easy-tool_selection-01": [
     "c",
     0.374
    ],
    "easy-tool_selection-02": [
     "c",
     0.361
    ],
    "easy-tool_selection-03": [
     "c",
     0.375
    ],
    "easy-tool_selection-04": [
     "c",
     0.374
    ],
    "easy-tool_selection-05": [
     "c",
     0.364
    ],
    "easy-tool_selection-06": [
     "c",
     0.372
    ],
    "easy-tool_selection-07": [
     "c",
     0.379
    ],
    "easy-tool_selection-08": [
     "c",
     0.366
    ],
    "easy-tool_selection-09": [
     "c",
     0.374
    ],
    "easy-tool_selection-10": [
     "c",
     0.365
    ],
    "easy-tool_selection-11": [
     "c",
     0.368
    ],
    "original-policy-01-0": [
     "c",
     0.356
    ],
    "original-policy-01-1": [
     "c",
     0.353
    ],
    "original-policy-02-0": [
     "c",
     0.354
    ],
    "original-policy-02-1": [
     "c",
     0.352
    ],
    "original-policy-03-0": [
     "c",
     0.349
    ],
    "original-policy-03-1": [
     "c",
     0.35
    ],
    "original-policy-04-0": [
     "c",
     0.356
    ],
    "original-policy-04-1": [
     "c",
     0.353
    ],
    "original-policy-05-0": [
     "c",
     0.357
    ],
    "original-policy-05-1": [
     "c",
     0.358
    ],
    "original-policy-06-0": [
     "c",
     0.352
    ],
    "original-policy-06-1": [
     "c",
     0.355
    ],
    "original-intent-01-0": [
     "c",
     0.371
    ],
    "original-intent-01-1": [
     "c",
     0.371
    ],
    "original-intent-02-0": [
     "c",
     0.362
    ],
    "original-intent-02-1": [
     "c",
     0.372
    ],
    "original-intent-03-0": [
     "c",
     0.366
    ],
    "original-intent-03-1": [
     "c",
     0.369
    ],
    "original-intent-04-0": [
     "c",
     0.373
    ],
    "original-intent-04-1": [
     "c",
     0.367
    ],
    "original-intent-05-0": [
     "w",
     0.373
    ],
    "original-intent-05-1": [
     "w",
     0.368
    ],
    "original-intent-06-0": [
     "c",
     0.364
    ],
    "original-intent-06-1": [
     "c",
     0.371
    ],
    "original-ordinal-01-0": [
     "c",
     0.364
    ],
    "original-ordinal-01-1": [
     "c",
     0.36
    ],
    "original-ordinal-02-0": [
     "c",
     0.364
    ],
    "original-ordinal-02-1": [
     "c",
     0.364
    ],
    "original-ordinal-03-0": [
     "c",
     0.367
    ],
    "original-ordinal-03-1": [
     "c",
     0.362
    ],
    "original-ordinal-04-0": [
     "c",
     0.358
    ],
    "original-ordinal-04-1": [
     "c",
     0.362
    ],
    "original-ordinal-05-0": [
     "c",
     0.367
    ],
    "original-ordinal-05-1": [
     "c",
     0.357
    ],
    "original-ordinal-06-0": [
     "c",
     0.387
    ],
    "original-ordinal-06-1": [
     "c",
     0.365
    ],
    "original-extraction-01-0": [
     "c",
     0.357
    ],
    "original-extraction-01-1": [
     "c",
     0.361
    ],
    "original-extraction-02-0": [
     "c",
     0.356
    ],
    "original-extraction-02-1": [
     "c",
     0.361
    ],
    "original-extraction-03-0": [
     "c",
     0.357
    ],
    "original-extraction-03-1": [
     "c",
     0.361
    ],
    "original-extraction-04-0": [
     "c",
     0.363
    ],
    "original-extraction-04-1": [
     "c",
     0.365
    ],
    "original-extraction-05-0": [
     "c",
     0.362
    ],
    "original-extraction-05-1": [
     "c",
     0.361
    ],
    "original-extraction-06-0": [
     "c",
     0.361
    ],
    "original-extraction-06-1": [
     "c",
     0.362
    ],
    "original-adequacy-01-0": [
     "c",
     0.36
    ],
    "original-adequacy-01-1": [
     "c",
     0.374
    ],
    "original-adequacy-02-0": [
     "c",
     0.376
    ],
    "original-adequacy-02-1": [
     "c",
     0.372
    ],
    "original-adequacy-03-0": [
     "w",
     0.37
    ],
    "original-adequacy-03-1": [
     "w",
     0.364
    ],
    "original-adequacy-04-0": [
     "c",
     0.363
    ],
    "original-adequacy-04-1": [
     "c",
     0.359
    ],
    "original-adequacy-05-0": [
     "w",
     0.361
    ],
    "original-adequacy-05-1": [
     "c",
     0.359
    ],
    "original-adequacy-06-0": [
     "c",
     0.361
    ],
    "original-adequacy-06-1": [
     "c",
     0.362
    ],
    "original-routing-01-0": [
     "c",
     0.371
    ],
    "original-routing-01-1": [
     "c",
     0.379
    ],
    "original-routing-02-0": [
     "c",
     0.378
    ],
    "original-routing-02-1": [
     "c",
     0.373
    ],
    "original-routing-03-0": [
     "c",
     0.389
    ],
    "original-routing-03-1": [
     "c",
     0.367
    ],
    "original-routing-04-0": [
     "c",
     0.373
    ],
    "original-routing-04-1": [
     "c",
     0.371
    ],
    "original-routing-05-0": [
     "c",
     0.382
    ],
    "original-routing-05-1": [
     "c",
     0.382
    ],
    "original-routing-06-0": [
     "c",
     0.376
    ],
    "original-routing-06-1": [
     "c",
     0.375
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     1.126
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.788
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.733
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.757
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.705
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     0.737
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.76
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.717
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.399
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.509
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.527
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.405
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.427
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.433
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.59
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.427
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.459
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.43
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.434
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.429
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.461
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.416
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.418
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.46
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     0.869
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     0.72
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.772
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.796
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.715
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.659
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.428
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.515
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.454
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.522
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.513
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.476
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.468
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.481
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.449
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.439
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     1.099
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     1.055
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.977
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.996
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.009
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     1.043
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.972
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.425
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.475
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     0.487
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.476
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.535
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.524
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.481
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.389
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.382
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     0.402
    ],
    "hard-sol-a-adversarial-08": [
     "w",
     0.383
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.381
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.397
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.772
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.72
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.717
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.727
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.727
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.719
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.731
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.733
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.38
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.376
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.409
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.398
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.397
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.379
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.37
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.383
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.376
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.369
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.378
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.372
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.373
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.387
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.381
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.369
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.892
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.887
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.87
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.888
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.397
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.396
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.386
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.383
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.394
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.395
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.397
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.392
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.391
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.406
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.404
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.389
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.391
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.398
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.404
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.4
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.394
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.386
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.799
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.756
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.756
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.439
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.758
    ]
   }
  },
  "openjev-sglang": {
   "display": "openjev-sglang (Qwen3.6-35B-A3B on SGLang)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 92,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9583
    },
    "judge": {
     "c": 139,
     "w": 7,
     "f": 0,
     "n": 0,
     "accuracy": 0.9521
    },
    "hard": {
     "c": 157,
     "w": 63,
     "f": 0,
     "n": 0,
     "accuracy": 0.7136
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 78,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 61,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8971
     }
    },
    "hard": {
     "long_policy": {
      "c": 23,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.6053
     },
     "probability": {
      "c": 11,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.55
     },
     "temporal_numeric": {
      "c": 12,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "ambiguous": {
      "c": 11,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.7857
     },
     "multi_hop": {
      "c": 28,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8
     },
     "tradeoff": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 27,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.8182
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.705
    ],
    "easy-intent-01": [
     "c",
     0.899
    ],
    "easy-intent-02": [
     "c",
     0.853
    ],
    "easy-intent-03": [
     "c",
     0.662
    ],
    "easy-intent-04": [
     "c",
     0.657
    ],
    "easy-intent-05": [
     "c",
     0.639
    ],
    "easy-intent-06": [
     "c",
     0.66
    ],
    "easy-intent-07": [
     "c",
     0.811
    ],
    "easy-intent-08": [
     "c",
     0.612
    ],
    "easy-intent-09": [
     "c",
     0.603
    ],
    "easy-intent-10": [
     "c",
     0.7
    ],
    "easy-intent-11": [
     "c",
     0.626
    ],
    "easy-fact-00": [
     "c",
     0.692
    ],
    "easy-fact-01": [
     "c",
     0.673
    ],
    "easy-fact-02": [
     "c",
     0.659
    ],
    "easy-fact-03": [
     "c",
     0.608
    ],
    "easy-fact-04": [
     "c",
     0.688
    ],
    "easy-fact-05": [
     "c",
     0.681
    ],
    "easy-fact-06": [
     "c",
     0.615
    ],
    "easy-fact-07": [
     "c",
     0.688
    ],
    "easy-fact-08": [
     "c",
     0.616
    ],
    "easy-fact-09": [
     "c",
     0.61
    ],
    "easy-fact-10": [
     "c",
     0.702
    ],
    "easy-fact-11": [
     "c",
     0.681
    ],
    "easy-extraction-00": [
     "c",
     0.64
    ],
    "easy-extraction-01": [
     "c",
     0.67
    ],
    "easy-extraction-02": [
     "c",
     0.683
    ],
    "easy-extraction-03": [
     "c",
     0.611
    ],
    "easy-extraction-04": [
     "c",
     0.679
    ],
    "easy-extraction-05": [
     "c",
     0.696
    ],
    "easy-extraction-06": [
     "c",
     0.603
    ],
    "easy-extraction-07": [
     "c",
     0.616
    ],
    "easy-extraction-08": [
     "c",
     0.685
    ],
    "easy-extraction-09": [
     "c",
     0.773
    ],
    "easy-extraction-10": [
     "c",
     0.694
    ],
    "easy-extraction-11": [
     "c",
     0.668
    ],
    "easy-tool_selection-00": [
     "c",
     0.711
    ],
    "easy-tool_selection-01": [
     "c",
     0.618
    ],
    "easy-tool_selection-02": [
     "c",
     0.6
    ],
    "easy-tool_selection-03": [
     "c",
     0.604
    ],
    "easy-tool_selection-04": [
     "c",
     0.595
    ],
    "easy-tool_selection-05": [
     "c",
     0.613
    ],
    "easy-tool_selection-06": [
     "c",
     0.691
    ],
    "easy-tool_selection-07": [
     "c",
     0.739
    ],
    "easy-tool_selection-08": [
     "c",
     0.691
    ],
    "easy-tool_selection-09": [
     "c",
     0.683
    ],
    "easy-tool_selection-10": [
     "c",
     0.696
    ],
    "easy-tool_selection-11": [
     "c",
     0.673
    ],
    "original-policy-01-0": [
     "c",
     0.851
    ],
    "original-policy-01-1": [
     "c",
     0.822
    ],
    "original-policy-02-0": [
     "c",
     0.752
    ],
    "original-policy-02-1": [
     "c",
     0.686
    ],
    "original-policy-03-0": [
     "w",
     0.679
    ],
    "original-policy-03-1": [
     "c",
     0.612
    ],
    "original-policy-04-0": [
     "c",
     1.009
    ],
    "original-policy-04-1": [
     "c",
     0.786
    ],
    "original-policy-05-0": [
     "c",
     0.697
    ],
    "original-policy-05-1": [
     "c",
     0.643
    ],
    "original-policy-06-0": [
     "c",
     0.689
    ],
    "original-policy-06-1": [
     "w",
     0.701
    ],
    "original-intent-01-0": [
     "c",
     0.688
    ],
    "original-intent-01-1": [
     "c",
     0.726
    ],
    "original-intent-02-0": [
     "c",
     0.62
    ],
    "original-intent-02-1": [
     "c",
     0.698
    ],
    "original-intent-03-0": [
     "c",
     0.599
    ],
    "original-intent-03-1": [
     "c",
     0.632
    ],
    "original-intent-04-0": [
     "c",
     0.686
    ],
    "original-intent-04-1": [
     "c",
     0.689
    ],
    "original-intent-05-0": [
     "c",
     0.697
    ],
    "original-intent-05-1": [
     "c",
     0.621
    ],
    "original-intent-06-0": [
     "c",
     0.691
    ],
    "original-intent-06-1": [
     "c",
     0.714
    ],
    "original-ordinal-01-0": [
     "c",
     0.721
    ],
    "original-ordinal-01-1": [
     "c",
     0.679
    ],
    "original-ordinal-02-0": [
     "c",
     0.687
    ],
    "original-ordinal-02-1": [
     "c",
     0.621
    ],
    "original-ordinal-03-0": [
     "c",
     0.685
    ],
    "original-ordinal-03-1": [
     "c",
     0.61
    ],
    "original-ordinal-04-0": [
     "c",
     0.65
    ],
    "original-ordinal-04-1": [
     "c",
     0.626
    ],
    "original-ordinal-05-0": [
     "c",
     0.698
    ],
    "original-ordinal-05-1": [
     "c",
     0.607
    ],
    "original-ordinal-06-0": [
     "c",
     0.648
    ],
    "original-ordinal-06-1": [
     "c",
     0.62
    ],
    "original-extraction-01-0": [
     "c",
     0.613
    ],
    "original-extraction-01-1": [
     "c",
     0.695
    ],
    "original-extraction-02-0": [
     "c",
     0.702
    ],
    "original-extraction-02-1": [
     "c",
     0.621
    ],
    "original-extraction-03-0": [
     "c",
     0.687
    ],
    "original-extraction-03-1": [
     "c",
     0.69
    ],
    "original-extraction-04-0": [
     "c",
     0.619
    ],
    "original-extraction-04-1": [
     "c",
     0.599
    ],
    "original-extraction-05-0": [
     "c",
     0.678
    ],
    "original-extraction-05-1": [
     "c",
     0.692
    ],
    "original-extraction-06-0": [
     "c",
     0.681
    ],
    "original-extraction-06-1": [
     "c",
     0.689
    ],
    "original-adequacy-01-0": [
     "c",
     0.688
    ],
    "original-adequacy-01-1": [
     "c",
     0.609
    ],
    "original-adequacy-02-0": [
     "c",
     0.61
    ],
    "original-adequacy-02-1": [
     "c",
     0.62
    ],
    "original-adequacy-03-0": [
     "c",
     0.692
    ],
    "original-adequacy-03-1": [
     "w",
     0.615
    ],
    "original-adequacy-04-0": [
     "c",
     0.713
    ],
    "original-adequacy-04-1": [
     "c",
     0.691
    ],
    "original-adequacy-05-0": [
     "w",
     0.606
    ],
    "original-adequacy-05-1": [
     "c",
     0.717
    ],
    "original-adequacy-06-0": [
     "c",
     0.686
    ],
    "original-adequacy-06-1": [
     "c",
     0.698
    ],
    "original-routing-01-0": [
     "c",
     0.691
    ],
    "original-routing-01-1": [
     "c",
     0.635
    ],
    "original-routing-02-0": [
     "c",
     0.617
    ],
    "original-routing-02-1": [
     "c",
     0.604
    ],
    "original-routing-03-0": [
     "c",
     0.68
    ],
    "original-routing-03-1": [
     "c",
     0.72
    ],
    "original-routing-04-0": [
     "c",
     0.683
    ],
    "original-routing-04-1": [
     "c",
     0.685
    ],
    "original-routing-05-0": [
     "c",
     0.698
    ],
    "original-routing-05-1": [
     "c",
     0.779
    ],
    "original-routing-06-0": [
     "c",
     0.622
    ],
    "original-routing-06-1": [
     "c",
     0.687
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.924
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.894
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.889
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.698
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.752
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.639
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.695
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.759
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.77
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.689
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.647
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.683
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.72
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.667
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     0.638
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.706
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.691
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.69
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.682
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.794
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.699
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.782
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.62
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.717
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     0.718
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.666
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.709
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.721
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.645
    ],
    "hard-opus-b-probability-01": [
     "w",
     0.714
    ],
    "hard-opus-b-probability-02": [
     "c",
     0.712
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.64
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.615
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.623
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.7
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.62
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.693
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.615
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.69
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.621
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.901
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.875
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.887
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.826
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.831
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.833
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.932
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.65
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.62
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     0.697
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.642
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.699
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.711
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.705
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.69
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.616
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.613
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.693
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.681
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.69
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.636
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.696
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.704
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.732
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.668
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.652
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.706
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.706
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.636
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.7
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.627
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.647
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.717
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.692
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.696
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.62
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.624
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.697
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.701
    ],
    "hard-sol-b-judge_hard-08": [
     "c",
     0.698
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.624
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.708
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.701
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.621
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.821
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.815
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.83
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.816
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.734
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.701
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.709
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.696
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.646
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.621
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.614
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.701
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.696
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.632
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.625
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.65
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.707
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.697
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.613
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.699
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.617
    ],
    "hard-sol-c-judge_hard-15": [
     "w",
     0.703
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.733
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.71
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.639
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.79
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.651
    ]
   }
  },
  "gpt-5.6-luna": {
   "display": "GPT-5.6 Luna (low reasoning effort)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 94,
     "w": 2,
     "f": 0,
     "n": 0,
     "accuracy": 0.9792
    },
    "judge": {
     "c": 141,
     "w": 5,
     "f": 0,
     "n": 0,
     "accuracy": 0.9658
    },
    "hard": {
     "c": 208,
     "w": 12,
     "f": 0,
     "n": 0,
     "accuracy": 0.9455
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 78,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 36,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.9474
     },
     "probability": {
      "c": 18,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.9
     },
     "temporal_numeric": {
      "c": 28,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.9333
     },
     "ambiguous": {
      "c": 13,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9286
     },
     "multi_hop": {
      "c": 33,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.9429
     },
     "tradeoff": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 30,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.9091
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.83
    ],
    "easy-intent-01": [
     "c",
     1.359
    ],
    "easy-intent-02": [
     "c",
     1.342
    ],
    "easy-intent-03": [
     "c",
     1.071
    ],
    "easy-intent-04": [
     "c",
     1.351
    ],
    "easy-intent-05": [
     "c",
     1.276
    ],
    "easy-intent-06": [
     "c",
     1.042
    ],
    "easy-intent-07": [
     "c",
     1.028
    ],
    "easy-intent-08": [
     "c",
     1.026
    ],
    "easy-intent-09": [
     "c",
     1.063
    ],
    "easy-intent-10": [
     "c",
     1.23
    ],
    "easy-intent-11": [
     "c",
     1.434
    ],
    "easy-fact-00": [
     "c",
     0.86
    ],
    "easy-fact-01": [
     "c",
     0.761
    ],
    "easy-fact-02": [
     "c",
     0.659
    ],
    "easy-fact-03": [
     "c",
     0.672
    ],
    "easy-fact-04": [
     "c",
     0.64
    ],
    "easy-fact-05": [
     "c",
     1.007
    ],
    "easy-fact-06": [
     "c",
     0.636
    ],
    "easy-fact-07": [
     "c",
     0.641
    ],
    "easy-fact-08": [
     "c",
     0.609
    ],
    "easy-fact-09": [
     "c",
     0.828
    ],
    "easy-fact-10": [
     "c",
     0.651
    ],
    "easy-fact-11": [
     "c",
     0.666
    ],
    "easy-extraction-00": [
     "c",
     1.41
    ],
    "easy-extraction-01": [
     "c",
     0.977
    ],
    "easy-extraction-02": [
     "c",
     0.768
    ],
    "easy-extraction-03": [
     "c",
     0.884
    ],
    "easy-extraction-04": [
     "c",
     1.006
    ],
    "easy-extraction-05": [
     "c",
     0.885
    ],
    "easy-extraction-06": [
     "c",
     0.819
    ],
    "easy-extraction-07": [
     "c",
     0.82
    ],
    "easy-extraction-08": [
     "c",
     0.774
    ],
    "easy-extraction-09": [
     "c",
     0.875
    ],
    "easy-extraction-10": [
     "c",
     0.873
    ],
    "easy-extraction-11": [
     "c",
     0.946
    ],
    "easy-tool_selection-00": [
     "c",
     0.855
    ],
    "easy-tool_selection-01": [
     "c",
     0.881
    ],
    "easy-tool_selection-02": [
     "c",
     0.952
    ],
    "easy-tool_selection-03": [
     "c",
     1.024
    ],
    "easy-tool_selection-04": [
     "c",
     0.924
    ],
    "easy-tool_selection-05": [
     "c",
     0.973
    ],
    "easy-tool_selection-06": [
     "c",
     1.199
    ],
    "easy-tool_selection-07": [
     "c",
     0.892
    ],
    "easy-tool_selection-08": [
     "c",
     1.232
    ],
    "easy-tool_selection-09": [
     "c",
     1.176
    ],
    "easy-tool_selection-10": [
     "c",
     1.236
    ],
    "easy-tool_selection-11": [
     "c",
     1.187
    ],
    "original-policy-01-0": [
     "c",
     0.786
    ],
    "original-policy-01-1": [
     "c",
     1.436
    ],
    "original-policy-02-0": [
     "c",
     0.882
    ],
    "original-policy-02-1": [
     "c",
     1.103
    ],
    "original-policy-03-0": [
     "c",
     1.063
    ],
    "original-policy-03-1": [
     "c",
     0.721
    ],
    "original-policy-04-0": [
     "c",
     0.699
    ],
    "original-policy-04-1": [
     "c",
     0.643
    ],
    "original-policy-05-0": [
     "c",
     0.868
    ],
    "original-policy-05-1": [
     "c",
     0.984
    ],
    "original-policy-06-0": [
     "c",
     0.612
    ],
    "original-policy-06-1": [
     "w",
     0.805
    ],
    "original-intent-01-0": [
     "c",
     1.312
    ],
    "original-intent-01-1": [
     "c",
     1.029
    ],
    "original-intent-02-0": [
     "c",
     0.984
    ],
    "original-intent-02-1": [
     "c",
     0.859
    ],
    "original-intent-03-0": [
     "c",
     0.999
    ],
    "original-intent-03-1": [
     "c",
     0.942
    ],
    "original-intent-04-0": [
     "c",
     1.029
    ],
    "original-intent-04-1": [
     "c",
     0.722
    ],
    "original-intent-05-0": [
     "c",
     1.167
    ],
    "original-intent-05-1": [
     "c",
     0.838
    ],
    "original-intent-06-0": [
     "c",
     1.022
    ],
    "original-intent-06-1": [
     "c",
     0.952
    ],
    "original-ordinal-01-0": [
     "c",
     0.851
    ],
    "original-ordinal-01-1": [
     "c",
     0.885
    ],
    "original-ordinal-02-0": [
     "c",
     0.775
    ],
    "original-ordinal-02-1": [
     "c",
     1.007
    ],
    "original-ordinal-03-0": [
     "c",
     0.865
    ],
    "original-ordinal-03-1": [
     "c",
     0.705
    ],
    "original-ordinal-04-0": [
     "c",
     0.885
    ],
    "original-ordinal-04-1": [
     "c",
     0.7
    ],
    "original-ordinal-05-0": [
     "c",
     0.725
    ],
    "original-ordinal-05-1": [
     "c",
     0.717
    ],
    "original-ordinal-06-0": [
     "c",
     0.948
    ],
    "original-ordinal-06-1": [
     "c",
     0.92
    ],
    "original-extraction-01-0": [
     "c",
     0.915
    ],
    "original-extraction-01-1": [
     "c",
     0.841
    ],
    "original-extraction-02-0": [
     "c",
     0.774
    ],
    "original-extraction-02-1": [
     "c",
     0.689
    ],
    "original-extraction-03-0": [
     "c",
     0.988
    ],
    "original-extraction-03-1": [
     "c",
     0.927
    ],
    "original-extraction-04-0": [
     "c",
     0.975
    ],
    "original-extraction-04-1": [
     "c",
     0.646
    ],
    "original-extraction-05-0": [
     "c",
     0.778
    ],
    "original-extraction-05-1": [
     "c",
     0.676
    ],
    "original-extraction-06-0": [
     "c",
     0.683
    ],
    "original-extraction-06-1": [
     "c",
     0.754
    ],
    "original-adequacy-01-0": [
     "c",
     0.756
    ],
    "original-adequacy-01-1": [
     "c",
     0.922
    ],
    "original-adequacy-02-0": [
     "c",
     0.837
    ],
    "original-adequacy-02-1": [
     "c",
     0.606
    ],
    "original-adequacy-03-0": [
     "c",
     0.661
    ],
    "original-adequacy-03-1": [
     "c",
     0.769
    ],
    "original-adequacy-04-0": [
     "c",
     0.8
    ],
    "original-adequacy-04-1": [
     "c",
     0.604
    ],
    "original-adequacy-05-0": [
     "c",
     0.973
    ],
    "original-adequacy-05-1": [
     "c",
     0.59
    ],
    "original-adequacy-06-0": [
     "c",
     0.671
    ],
    "original-adequacy-06-1": [
     "w",
     0.628
    ],
    "original-routing-01-0": [
     "c",
     0.929
    ],
    "original-routing-01-1": [
     "c",
     0.701
    ],
    "original-routing-02-0": [
     "c",
     0.974
    ],
    "original-routing-02-1": [
     "c",
     0.747
    ],
    "original-routing-03-0": [
     "c",
     0.825
    ],
    "original-routing-03-1": [
     "c",
     0.69
    ],
    "original-routing-04-0": [
     "c",
     0.85
    ],
    "original-routing-04-1": [
     "c",
     0.738
    ],
    "original-routing-05-0": [
     "c",
     0.86
    ],
    "original-routing-05-1": [
     "c",
     0.662
    ],
    "original-routing-06-0": [
     "c",
     1.0
    ],
    "original-routing-06-1": [
     "c",
     0.946
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     2.04
    ],
    "hard-opus-a-long_policy-04": [
     "c",
     1.939
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     1.729
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     1.889
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.905
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     1.851
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     1.228
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     1.443
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.297
    ],
    "hard-opus-a-probability-04": [
     "c",
     2.72
    ],
    "hard-opus-a-probability-07": [
     "c",
     2.032
    ],
    "hard-opus-a-probability-08": [
     "c",
     1.384
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     1.742
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     1.242
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     2.438
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.877
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     2.654
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     1.692
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     1.694
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.888
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.831
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.945
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     1.272
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     1.007
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     1.829
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     1.63
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     1.417
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     1.472
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     1.786
    ],
    "hard-opus-b-probability-01": [
     "c",
     2.509
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.244
    ],
    "hard-opus-b-probability-03": [
     "c",
     1.752
    ],
    "hard-opus-b-probability-04": [
     "c",
     1.346
    ],
    "hard-opus-b-probability-06": [
     "c",
     1.896
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     1.125
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     1.225
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     1.09
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     1.14
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     1.134
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     1.219
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.902
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     2.121
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.777
    ],
    "hard-opus-c-long_policy-05": [
     "c",
     3.286
    ],
    "hard-opus-c-long_policy-08": [
     "c",
     1.769
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     1.799
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     3.4
    ],
    "hard-opus-c-probability-03": [
     "c",
     1.975
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     2.92
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     2.222
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     4.973
    ],
    "hard-opus-c-temporal_numeric-06": [
     "c",
     2.53
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     1.972
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     2.252
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.858
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.694
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.158
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.825
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     1.114
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     1.042
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.893
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     1.818
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     1.725
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     1.222
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     1.032
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     1.456
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     1.167
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     1.247
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.775
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.751
    ],
    "hard-sol-a-trap-05": [
     "c",
     1.054
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.946
    ],
    "hard-sol-a-trap-08": [
     "c",
     1.273
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.969
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.721
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.95
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.659
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.77
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.624
    ],
    "hard-sol-b-judge_hard-08": [
     "c",
     0.873
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     1.15
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.828
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.74
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.781
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     1.868
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.948
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.378
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     1.558
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.918
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.031
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.992
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     1.223
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.147
    ],
    "hard-sol-b-temporal_numeric-01": [
     "c",
     1.442
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     1.941
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     1.216
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     1.199
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.774
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     1.102
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.859
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     1.143
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.673
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     1.491
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.665
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     1.122
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.874
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     1.5
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.831
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     1.269
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.882
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.815
    ]
   }
  },
  "gemini-3.1-flash-lite": {
   "display": "Gemini 3.1 Flash-Lite",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 95,
     "w": 1,
     "f": 0,
     "n": 0,
     "accuracy": 0.9896
    },
    "judge": {
     "c": 136,
     "w": 10,
     "f": 0,
     "n": 0,
     "accuracy": 0.9315
    },
    "hard": {
     "c": 165,
     "w": 55,
     "f": 0,
     "n": 0,
     "accuracy": 0.75
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 76,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.9744
     },
     "adequacy": {
      "c": 60,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8824
     }
    },
    "hard": {
     "long_policy": {
      "c": 25,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.6579
     },
     "probability": {
      "c": 18,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.9
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 11,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.7857
     },
     "multi_hop": {
      "c": 27,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.7714
     },
     "tradeoff": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 29,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.8788
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.899
    ],
    "easy-intent-01": [
     "c",
     1.392
    ],
    "easy-intent-02": [
     "c",
     0.995
    ],
    "easy-intent-03": [
     "c",
     0.752
    ],
    "easy-intent-04": [
     "c",
     0.826
    ],
    "easy-intent-05": [
     "c",
     0.919
    ],
    "easy-intent-06": [
     "c",
     0.841
    ],
    "easy-intent-07": [
     "c",
     0.858
    ],
    "easy-intent-08": [
     "c",
     0.87
    ],
    "easy-intent-09": [
     "c",
     0.843
    ],
    "easy-intent-10": [
     "c",
     0.823
    ],
    "easy-intent-11": [
     "c",
     0.806
    ],
    "easy-fact-00": [
     "c",
     0.699
    ],
    "easy-fact-01": [
     "c",
     0.69
    ],
    "easy-fact-02": [
     "c",
     0.788
    ],
    "easy-fact-03": [
     "c",
     0.627
    ],
    "easy-fact-04": [
     "c",
     0.682
    ],
    "easy-fact-05": [
     "c",
     0.682
    ],
    "easy-fact-06": [
     "c",
     0.665
    ],
    "easy-fact-07": [
     "c",
     0.766
    ],
    "easy-fact-08": [
     "c",
     0.746
    ],
    "easy-fact-09": [
     "c",
     0.682
    ],
    "easy-fact-10": [
     "c",
     0.728
    ],
    "easy-fact-11": [
     "c",
     0.657
    ],
    "easy-extraction-00": [
     "c",
     0.883
    ],
    "easy-extraction-01": [
     "c",
     0.786
    ],
    "easy-extraction-02": [
     "c",
     0.8
    ],
    "easy-extraction-03": [
     "c",
     0.717
    ],
    "easy-extraction-04": [
     "c",
     0.931
    ],
    "easy-extraction-05": [
     "c",
     0.777
    ],
    "easy-extraction-06": [
     "c",
     0.781
    ],
    "easy-extraction-07": [
     "c",
     0.788
    ],
    "easy-extraction-08": [
     "c",
     0.797
    ],
    "easy-extraction-09": [
     "c",
     0.773
    ],
    "easy-extraction-10": [
     "c",
     0.778
    ],
    "easy-extraction-11": [
     "c",
     0.721
    ],
    "easy-tool_selection-00": [
     "c",
     0.752
    ],
    "easy-tool_selection-01": [
     "c",
     0.754
    ],
    "easy-tool_selection-02": [
     "c",
     0.788
    ],
    "easy-tool_selection-03": [
     "c",
     0.856
    ],
    "easy-tool_selection-04": [
     "c",
     0.843
    ],
    "easy-tool_selection-05": [
     "c",
     1.155
    ],
    "easy-tool_selection-06": [
     "c",
     0.838
    ],
    "easy-tool_selection-07": [
     "c",
     0.742
    ],
    "easy-tool_selection-08": [
     "c",
     0.894
    ],
    "easy-tool_selection-09": [
     "c",
     0.816
    ],
    "easy-tool_selection-10": [
     "c",
     0.768
    ],
    "easy-tool_selection-11": [
     "c",
     0.862
    ],
    "original-policy-01-0": [
     "c",
     0.732
    ],
    "original-policy-01-1": [
     "c",
     0.72
    ],
    "original-policy-02-0": [
     "c",
     0.667
    ],
    "original-policy-02-1": [
     "c",
     0.686
    ],
    "original-policy-03-0": [
     "c",
     0.751
    ],
    "original-policy-03-1": [
     "c",
     0.774
    ],
    "original-policy-04-0": [
     "c",
     0.647
    ],
    "original-policy-04-1": [
     "c",
     0.664
    ],
    "original-policy-05-0": [
     "c",
     0.647
    ],
    "original-policy-05-1": [
     "c",
     0.769
    ],
    "original-policy-06-0": [
     "c",
     0.736
    ],
    "original-policy-06-1": [
     "c",
     0.69
    ],
    "original-intent-01-0": [
     "c",
     0.768
    ],
    "original-intent-01-1": [
     "c",
     0.805
    ],
    "original-intent-02-0": [
     "c",
     0.819
    ],
    "original-intent-02-1": [
     "c",
     0.791
    ],
    "original-intent-03-0": [
     "c",
     0.718
    ],
    "original-intent-03-1": [
     "c",
     0.766
    ],
    "original-intent-04-0": [
     "c",
     0.796
    ],
    "original-intent-04-1": [
     "c",
     0.735
    ],
    "original-intent-05-0": [
     "c",
     0.802
    ],
    "original-intent-05-1": [
     "c",
     0.676
    ],
    "original-intent-06-0": [
     "c",
     0.862
    ],
    "original-intent-06-1": [
     "c",
     0.771
    ],
    "original-ordinal-01-0": [
     "c",
     0.818
    ],
    "original-ordinal-01-1": [
     "c",
     0.978
    ],
    "original-ordinal-02-0": [
     "c",
     0.726
    ],
    "original-ordinal-02-1": [
     "c",
     0.843
    ],
    "original-ordinal-03-0": [
     "c",
     0.766
    ],
    "original-ordinal-03-1": [
     "c",
     0.74
    ],
    "original-ordinal-04-0": [
     "c",
     0.768
    ],
    "original-ordinal-04-1": [
     "c",
     0.717
    ],
    "original-ordinal-05-0": [
     "c",
     0.771
    ],
    "original-ordinal-05-1": [
     "c",
     0.792
    ],
    "original-ordinal-06-0": [
     "c",
     0.793
    ],
    "original-ordinal-06-1": [
     "c",
     0.714
    ],
    "original-extraction-01-0": [
     "c",
     0.79
    ],
    "original-extraction-01-1": [
     "c",
     0.742
    ],
    "original-extraction-02-0": [
     "c",
     0.684
    ],
    "original-extraction-02-1": [
     "c",
     0.744
    ],
    "original-extraction-03-0": [
     "c",
     0.727
    ],
    "original-extraction-03-1": [
     "c",
     0.762
    ],
    "original-extraction-04-0": [
     "c",
     0.733
    ],
    "original-extraction-04-1": [
     "c",
     0.804
    ],
    "original-extraction-05-0": [
     "c",
     0.706
    ],
    "original-extraction-05-1": [
     "c",
     0.776
    ],
    "original-extraction-06-0": [
     "c",
     0.725
    ],
    "original-extraction-06-1": [
     "c",
     0.814
    ],
    "original-adequacy-01-0": [
     "c",
     0.639
    ],
    "original-adequacy-01-1": [
     "c",
     0.67
    ],
    "original-adequacy-02-0": [
     "c",
     0.959
    ],
    "original-adequacy-02-1": [
     "c",
     0.638
    ],
    "original-adequacy-03-0": [
     "w",
     0.622
    ],
    "original-adequacy-03-1": [
     "c",
     0.697
    ],
    "original-adequacy-04-0": [
     "c",
     0.75
    ],
    "original-adequacy-04-1": [
     "c",
     1.189
    ],
    "original-adequacy-05-0": [
     "c",
     0.726
    ],
    "original-adequacy-05-1": [
     "c",
     2.472
    ],
    "original-adequacy-06-0": [
     "c",
     0.759
    ],
    "original-adequacy-06-1": [
     "c",
     0.812
    ],
    "original-routing-01-0": [
     "c",
     0.75
    ],
    "original-routing-01-1": [
     "c",
     0.81
    ],
    "original-routing-02-0": [
     "c",
     0.766
    ],
    "original-routing-02-1": [
     "c",
     1.116
    ],
    "original-routing-03-0": [
     "c",
     0.808
    ],
    "original-routing-03-1": [
     "c",
     0.763
    ],
    "original-routing-04-0": [
     "c",
     0.804
    ],
    "original-routing-04-1": [
     "c",
     0.736
    ],
    "original-routing-05-0": [
     "c",
     0.705
    ],
    "original-routing-05-1": [
     "c",
     0.76
    ],
    "original-routing-06-0": [
     "c",
     0.833
    ],
    "original-routing-06-1": [
     "c",
     0.816
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     1.188
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.859
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.871
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.942
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.792
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     1.126
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     1.061
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.889
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.777
    ],
    "hard-opus-a-probability-04": [
     "c",
     0.858
    ],
    "hard-opus-a-probability-07": [
     "c",
     0.978
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.873
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.752
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.862
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     1.643
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.855
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.637
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.665
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.834
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.936
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.815
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.814
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.859
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.768
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.676
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.812
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.645
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.876
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.84
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.85
    ],
    "hard-opus-b-probability-02": [
     "c",
     0.611
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.582
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.602
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.694
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.746
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.825
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.821
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.894
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.682
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.819
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     1.932
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.833
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.609
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.653
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.741
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.875
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     0.611
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.891
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.654
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.75
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.715
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.871
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     1.397
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.934
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.619
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.651
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.738
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.597
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.706
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.914
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.761
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.948
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.958
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     1.408
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.65
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.597
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.836
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.758
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.73
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.684
    ],
    "hard-sol-a-trap-05": [
     "c",
     1.313
    ],
    "hard-sol-a-trap-06": [
     "c",
     1.291
    ],
    "hard-sol-a-trap-08": [
     "c",
     1.387
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.728
    ],
    "hard-sol-a-trap-13": [
     "c",
     1.544
    ],
    "hard-sol-a-trap-15": [
     "c",
     1.081
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.883
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.558
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.759
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.516
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.818
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.663
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.599
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.801
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.964
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.69
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.949
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.714
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.666
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.718
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.758
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.679
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.796
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.16
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.812
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.705
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.685
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.698
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.539
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.671
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.552
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.762
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.838
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.628
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.7
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.548
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.81
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.616
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.734
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.68
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.678
    ]
   }
  },
  "open-jev-deberta-v3-large": {
   "display": "open-jev-deberta-v3-large (local CPU)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 47,
     "w": 49,
     "f": 0,
     "n": 0,
     "accuracy": 0.4896
    },
    "judge": {
     "c": 78,
     "w": 68,
     "f": 0,
     "n": 0,
     "accuracy": 0.5342
    },
    "hard": {
     "c": 80,
     "w": 138,
     "f": 2,
     "n": 0,
     "accuracy": 0.3636
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "ordinal": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "extraction": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "adequacy": {
      "c": 5,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.3125
     },
     "routing": {
      "c": 6,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.375
     }
    },
    "judge": {
     "routing": {
      "c": 18,
      "w": 60,
      "f": 0,
      "n": 0,
      "accuracy": 0.2308
     },
     "adequacy": {
      "c": 60,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8824
     }
    },
    "hard": {
     "long_policy": {
      "c": 13,
      "w": 23,
      "f": 2,
      "n": 0,
      "accuracy": 0.3421
     },
     "probability": {
      "c": 6,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "temporal_numeric": {
      "c": 4,
      "w": 26,
      "f": 0,
      "n": 0,
      "accuracy": 0.1333
     },
     "ambiguous": {
      "c": 6,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.4286
     },
     "multi_hop": {
      "c": 13,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.3714
     },
     "tradeoff": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 5,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.3125
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 2,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     8.054
    ],
    "easy-intent-01": [
     "c",
     1.343
    ],
    "easy-intent-02": [
     "c",
     1.233
    ],
    "easy-intent-03": [
     "c",
     1.209
    ],
    "easy-intent-04": [
     "c",
     1.24
    ],
    "easy-intent-05": [
     "c",
     1.17
    ],
    "easy-intent-06": [
     "c",
     1.217
    ],
    "easy-intent-07": [
     "c",
     1.137
    ],
    "easy-intent-08": [
     "c",
     1.178
    ],
    "easy-intent-09": [
     "c",
     1.119
    ],
    "easy-intent-10": [
     "c",
     1.056
    ],
    "easy-intent-11": [
     "c",
     1.033
    ],
    "easy-fact-00": [
     "c",
     0.827
    ],
    "easy-fact-01": [
     "c",
     0.814
    ],
    "easy-fact-02": [
     "c",
     0.817
    ],
    "easy-fact-03": [
     "c",
     0.852
    ],
    "easy-fact-04": [
     "c",
     0.821
    ],
    "easy-fact-05": [
     "c",
     0.83
    ],
    "easy-fact-06": [
     "c",
     0.869
    ],
    "easy-fact-07": [
     "c",
     0.828
    ],
    "easy-fact-08": [
     "c",
     0.818
    ],
    "easy-fact-09": [
     "c",
     0.83
    ],
    "easy-fact-10": [
     "c",
     0.831
    ],
    "easy-fact-11": [
     "c",
     0.839
    ],
    "easy-extraction-00": [
     "c",
     0.842
    ],
    "easy-extraction-01": [
     "c",
     0.901
    ],
    "easy-extraction-02": [
     "c",
     1.13
    ],
    "easy-extraction-03": [
     "c",
     0.825
    ],
    "easy-extraction-04": [
     "c",
     0.936
    ],
    "easy-extraction-05": [
     "c",
     0.867
    ],
    "easy-extraction-06": [
     "c",
     0.829
    ],
    "easy-extraction-07": [
     "c",
     0.809
    ],
    "easy-extraction-08": [
     "c",
     0.91
    ],
    "easy-extraction-09": [
     "c",
     1.033
    ],
    "easy-extraction-10": [
     "c",
     0.922
    ],
    "easy-extraction-11": [
     "c",
     0.922
    ],
    "easy-tool_selection-00": [
     "c",
     1.127
    ],
    "easy-tool_selection-01": [
     "c",
     1.107
    ],
    "easy-tool_selection-02": [
     "c",
     1.108
    ],
    "easy-tool_selection-03": [
     "c",
     1.252
    ],
    "easy-tool_selection-04": [
     "c",
     1.139
    ],
    "easy-tool_selection-05": [
     "c",
     1.213
    ],
    "easy-tool_selection-06": [
     "c",
     1.135
    ],
    "easy-tool_selection-07": [
     "c",
     1.132
    ],
    "easy-tool_selection-08": [
     "c",
     1.305
    ],
    "easy-tool_selection-09": [
     "c",
     1.431
    ],
    "easy-tool_selection-10": [
     "c",
     1.273
    ],
    "easy-tool_selection-11": [
     "c",
     1.244
    ],
    "original-policy-01-0": [
     "c",
     5.622
    ],
    "original-policy-01-1": [
     "c",
     0.92
    ],
    "original-policy-02-0": [
     "w",
     0.982
    ],
    "original-policy-02-1": [
     "w",
     0.99
    ],
    "original-policy-03-0": [
     "w",
     0.943
    ],
    "original-policy-03-1": [
     "w",
     0.946
    ],
    "original-policy-04-0": [
     "c",
     0.953
    ],
    "original-policy-04-1": [
     "c",
     0.915
    ],
    "original-policy-05-0": [
     "c",
     0.966
    ],
    "original-policy-05-1": [
     "c",
     1.0
    ],
    "original-policy-06-0": [
     "w",
     0.988
    ],
    "original-policy-06-1": [
     "w",
     0.89
    ],
    "original-intent-01-0": [
     "w",
     1.041
    ],
    "original-intent-01-1": [
     "c",
     1.01
    ],
    "original-intent-02-0": [
     "w",
     1.018
    ],
    "original-intent-02-1": [
     "c",
     1.047
    ],
    "original-intent-03-0": [
     "c",
     1.105
    ],
    "original-intent-03-1": [
     "c",
     1.051
    ],
    "original-intent-04-0": [
     "c",
     1.09
    ],
    "original-intent-04-1": [
     "c",
     1.048
    ],
    "original-intent-05-0": [
     "w",
     1.032
    ],
    "original-intent-05-1": [
     "w",
     1.197
    ],
    "original-intent-06-0": [
     "w",
     1.064
    ],
    "original-intent-06-1": [
     "w",
     1.098
    ],
    "original-ordinal-01-0": [
     "w",
     1.019
    ],
    "original-ordinal-01-1": [
     "w",
     1.072
    ],
    "original-ordinal-02-0": [
     "w",
     1.068
    ],
    "original-ordinal-02-1": [
     "w",
     1.166
    ],
    "original-ordinal-03-0": [
     "c",
     1.042
    ],
    "original-ordinal-03-1": [
     "c",
     1.19
    ],
    "original-ordinal-04-0": [
     "w",
     0.978
    ],
    "original-ordinal-04-1": [
     "c",
     1.047
    ],
    "original-ordinal-05-0": [
     "w",
     1.081
    ],
    "original-ordinal-05-1": [
     "w",
     1.098
    ],
    "original-ordinal-06-0": [
     "c",
     1.083
    ],
    "original-ordinal-06-1": [
     "c",
     1.124
    ],
    "original-extraction-01-0": [
     "w",
     0.964
    ],
    "original-extraction-01-1": [
     "c",
     0.932
    ],
    "original-extraction-02-0": [
     "w",
     0.988
    ],
    "original-extraction-02-1": [
     "w",
     0.941
    ],
    "original-extraction-03-0": [
     "c",
     0.954
    ],
    "original-extraction-03-1": [
     "c",
     0.954
    ],
    "original-extraction-04-0": [
     "c",
     0.904
    ],
    "original-extraction-04-1": [
     "c",
     0.918
    ],
    "original-extraction-05-0": [
     "w",
     0.97
    ],
    "original-extraction-05-1": [
     "w",
     1.011
    ],
    "original-extraction-06-0": [
     "c",
     1.0
    ],
    "original-extraction-06-1": [
     "c",
     0.963
    ],
    "original-adequacy-01-0": [
     "w",
     0.841
    ],
    "original-adequacy-01-1": [
     "w",
     0.838
    ],
    "original-adequacy-02-0": [
     "c",
     0.838
    ],
    "original-adequacy-02-1": [
     "c",
     0.919
    ],
    "original-adequacy-03-0": [
     "w",
     1.256
    ],
    "original-adequacy-03-1": [
     "w",
     1.255
    ],
    "original-adequacy-04-0": [
     "w",
     1.38
    ],
    "original-adequacy-04-1": [
     "w",
     1.129
    ],
    "original-adequacy-05-0": [
     "c",
     0.889
    ],
    "original-adequacy-05-1": [
     "w",
     0.875
    ],
    "original-adequacy-06-0": [
     "w",
     1.031
    ],
    "original-adequacy-06-1": [
     "w",
     0.966
    ],
    "original-routing-01-0": [
     "w",
     1.215
    ],
    "original-routing-01-1": [
     "w",
     1.346
    ],
    "original-routing-02-0": [
     "c",
     1.334
    ],
    "original-routing-02-1": [
     "c",
     1.279
    ],
    "original-routing-03-0": [
     "c",
     1.341
    ],
    "original-routing-03-1": [
     "c",
     1.284
    ],
    "original-routing-04-0": [
     "w",
     1.184
    ],
    "original-routing-04-1": [
     "w",
     1.155
    ],
    "original-routing-05-0": [
     "w",
     1.145
    ],
    "original-routing-05-1": [
     "w",
     1.148
    ],
    "original-routing-06-0": [
     "w",
     1.29
    ],
    "original-routing-06-1": [
     "w",
     1.357
    ],
    "hard-opus-a-long_policy-01": [
     "f",
     6.036
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     5.156
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     3.319
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     3.063
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     2.471
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     2.498
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     3.017
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     3.926
    ],
    "hard-opus-a-probability-03": [
     "c",
     2.576
    ],
    "hard-opus-a-probability-04": [
     "w",
     2.722
    ],
    "hard-opus-a-probability-07": [
     "w",
     2.586
    ],
    "hard-opus-a-probability-08": [
     "w",
     2.192
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     2.382
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     2.435
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     2.783
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     3.02
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     2.81
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     3.12
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     5.561
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     2.971
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     2.429
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     2.682
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     2.465
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     4.666
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     3.127
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     3.824
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     2.488
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     3.201
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     2.534
    ],
    "hard-opus-b-probability-01": [
     "w",
     2.826
    ],
    "hard-opus-b-probability-02": [
     "w",
     2.91
    ],
    "hard-opus-b-probability-03": [
     "c",
     2.473
    ],
    "hard-opus-b-probability-04": [
     "c",
     2.83
    ],
    "hard-opus-b-probability-06": [
     "w",
     2.407
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     3.463
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     3.195
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     2.51
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     3.94
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     2.756
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     2.924
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     3.907
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     3.225
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     2.527
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     2.762
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     2.692
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     3.256
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     3.13
    ],
    "hard-opus-c-probability-03": [
     "w",
     3.07
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     2.499
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     3.139
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     3.266
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     3.582
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     3.34
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     2.669
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     2.3
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     1.801
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     2.277
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     1.723
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     1.886
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     2.365
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     2.642
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     2.68
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     2.573
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     2.473
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     2.603
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     2.629
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     3.009
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     2.78
    ],
    "hard-sol-a-trap-02": [
     "w",
     1.892
    ],
    "hard-sol-a-trap-04": [
     "c",
     1.514
    ],
    "hard-sol-a-trap-05": [
     "w",
     2.268
    ],
    "hard-sol-a-trap-06": [
     "c",
     2.304
    ],
    "hard-sol-a-trap-08": [
     "w",
     1.956
    ],
    "hard-sol-a-trap-10": [
     "w",
     2.015
    ],
    "hard-sol-a-trap-13": [
     "c",
     1.718
    ],
    "hard-sol-a-trap-15": [
     "w",
     2.347
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.848
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     1.515
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     1.905
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     1.66
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     1.588
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     1.683
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     1.349
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     1.283
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     3.141
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     2.789
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     3.14
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     2.675
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     2.448
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     2.232
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     2.293
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     2.31
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     2.184
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.927
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     2.015
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     1.892
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     1.992
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     2.373
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     2.221
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     2.063
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     2.284
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     2.055
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     2.12
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     2.322
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     1.982
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     2.091
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     2.961
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     2.846
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     2.875
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     3.039
    ],
    "hard-sol-c-multi_hop-13": [
     "w",
     2.79
    ]
   }
  },
  "deepseek-flash": {
   "display": "DeepSeek V4.1 Flash (thinking default)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 71,
     "w": 0,
     "f": 1,
     "n": 0,
     "accuracy": 0.9861
    },
    "standard": {
     "c": 95,
     "w": 0,
     "f": 1,
     "n": 0,
     "accuracy": 0.9896
    },
    "judge": {
     "c": 136,
     "w": 3,
     "f": 7,
     "n": 0,
     "accuracy": 0.9315
    },
    "hard": {
     "c": 209,
     "w": 3,
     "f": 8,
     "n": 0,
     "accuracy": 0.95
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 17,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.9444
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 77,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.9872
     },
     "adequacy": {
      "c": 59,
      "w": 3,
      "f": 6,
      "n": 0,
      "accuracy": 0.8676
     }
    },
    "hard": {
     "long_policy": {
      "c": 34,
      "w": 0,
      "f": 4,
      "n": 0,
      "accuracy": 0.8947
     },
     "probability": {
      "c": 19,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.95
     },
     "temporal_numeric": {
      "c": 28,
      "w": 0,
      "f": 2,
      "n": 0,
      "accuracy": 0.9333
     },
     "ambiguous": {
      "c": 13,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.9286
     },
     "multi_hop": {
      "c": 35,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tradeoff": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 30,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.9091
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.582
    ],
    "easy-intent-01": [
     "c",
     1.098
    ],
    "easy-intent-02": [
     "c",
     1.119
    ],
    "easy-intent-03": [
     "c",
     0.764
    ],
    "easy-intent-04": [
     "c",
     0.984
    ],
    "easy-intent-05": [
     "c",
     0.872
    ],
    "easy-intent-06": [
     "c",
     1.811
    ],
    "easy-intent-07": [
     "c",
     1.686
    ],
    "easy-intent-08": [
     "c",
     1.082
    ],
    "easy-intent-09": [
     "c",
     1.143
    ],
    "easy-intent-10": [
     "c",
     1.198
    ],
    "easy-intent-11": [
     "c",
     3.939
    ],
    "easy-fact-00": [
     "c",
     1.151
    ],
    "easy-fact-01": [
     "c",
     0.968
    ],
    "easy-fact-02": [
     "c",
     1.329
    ],
    "easy-fact-03": [
     "c",
     0.722
    ],
    "easy-fact-04": [
     "c",
     1.309
    ],
    "easy-fact-05": [
     "c",
     0.856
    ],
    "easy-fact-06": [
     "c",
     0.702
    ],
    "easy-fact-07": [
     "c",
     0.941
    ],
    "easy-fact-08": [
     "c",
     1.055
    ],
    "easy-fact-09": [
     "c",
     0.925
    ],
    "easy-fact-10": [
     "c",
     0.91
    ],
    "easy-fact-11": [
     "c",
     1.138
    ],
    "easy-extraction-00": [
     "c",
     1.479
    ],
    "easy-extraction-01": [
     "c",
     1.257
    ],
    "easy-extraction-02": [
     "c",
     1.283
    ],
    "easy-extraction-03": [
     "c",
     1.224
    ],
    "easy-extraction-04": [
     "c",
     0.801
    ],
    "easy-extraction-05": [
     "c",
     0.974
    ],
    "easy-extraction-06": [
     "c",
     1.269
    ],
    "easy-extraction-07": [
     "c",
     1.403
    ],
    "easy-extraction-08": [
     "c",
     0.932
    ],
    "easy-extraction-09": [
     "c",
     0.922
    ],
    "easy-extraction-10": [
     "c",
     1.332
    ],
    "easy-extraction-11": [
     "c",
     0.823
    ],
    "easy-tool_selection-00": [
     "c",
     1.223
    ],
    "easy-tool_selection-01": [
     "c",
     1.155
    ],
    "easy-tool_selection-02": [
     "c",
     0.82
    ],
    "easy-tool_selection-03": [
     "c",
     1.139
    ],
    "easy-tool_selection-04": [
     "c",
     0.862
    ],
    "easy-tool_selection-05": [
     "c",
     1.599
    ],
    "easy-tool_selection-06": [
     "c",
     2.122
    ],
    "easy-tool_selection-07": [
     "c",
     0.937
    ],
    "easy-tool_selection-08": [
     "c",
     1.011
    ],
    "easy-tool_selection-09": [
     "c",
     0.882
    ],
    "easy-tool_selection-10": [
     "c",
     1.032
    ],
    "easy-tool_selection-11": [
     "c",
     0.633
    ],
    "original-policy-01-0": [
     "c",
     1.089
    ],
    "original-policy-01-1": [
     "f",
     3.215
    ],
    "original-policy-02-0": [
     "c",
     1.143
    ],
    "original-policy-02-1": [
     "c",
     1.104
    ],
    "original-policy-03-0": [
     "c",
     1.912
    ],
    "original-policy-03-1": [
     "c",
     3.542
    ],
    "original-policy-04-0": [
     "c",
     1.317
    ],
    "original-policy-04-1": [
     "c",
     1.289
    ],
    "original-policy-05-0": [
     "c",
     1.014
    ],
    "original-policy-05-1": [
     "c",
     0.968
    ],
    "original-policy-06-0": [
     "c",
     1.132
    ],
    "original-policy-06-1": [
     "c",
     1.91
    ],
    "original-intent-01-0": [
     "c",
     1.156
    ],
    "original-intent-01-1": [
     "c",
     0.901
    ],
    "original-intent-02-0": [
     "c",
     1.382
    ],
    "original-intent-02-1": [
     "c",
     1.075
    ],
    "original-intent-03-0": [
     "c",
     1.344
    ],
    "original-intent-03-1": [
     "c",
     1.157
    ],
    "original-intent-04-0": [
     "c",
     1.099
    ],
    "original-intent-04-1": [
     "c",
     1.011
    ],
    "original-intent-05-0": [
     "c",
     0.784
    ],
    "original-intent-05-1": [
     "c",
     0.954
    ],
    "original-intent-06-0": [
     "c",
     1.584
    ],
    "original-intent-06-1": [
     "c",
     1.044
    ],
    "original-ordinal-01-0": [
     "c",
     1.421
    ],
    "original-ordinal-01-1": [
     "c",
     1.501
    ],
    "original-ordinal-02-0": [
     "c",
     1.519
    ],
    "original-ordinal-02-1": [
     "c",
     1.249
    ],
    "original-ordinal-03-0": [
     "c",
     1.73
    ],
    "original-ordinal-03-1": [
     "c",
     1.951
    ],
    "original-ordinal-04-0": [
     "c",
     1.026
    ],
    "original-ordinal-04-1": [
     "c",
     1.311
    ],
    "original-ordinal-05-0": [
     "c",
     1.543
    ],
    "original-ordinal-05-1": [
     "c",
     2.117
    ],
    "original-ordinal-06-0": [
     "c",
     2.306
    ],
    "original-ordinal-06-1": [
     "c",
     1.422
    ],
    "original-extraction-01-0": [
     "c",
     0.794
    ],
    "original-extraction-01-1": [
     "c",
     0.937
    ],
    "original-extraction-02-0": [
     "c",
     1.122
    ],
    "original-extraction-02-1": [
     "c",
     1.13
    ],
    "original-extraction-03-0": [
     "c",
     0.941
    ],
    "original-extraction-03-1": [
     "c",
     0.833
    ],
    "original-extraction-04-0": [
     "c",
     0.947
    ],
    "original-extraction-04-1": [
     "c",
     1.197
    ],
    "original-extraction-05-0": [
     "c",
     1.346
    ],
    "original-extraction-05-1": [
     "c",
     0.937
    ],
    "original-extraction-06-0": [
     "c",
     1.124
    ],
    "original-extraction-06-1": [
     "c",
     1.049
    ],
    "original-adequacy-01-0": [
     "c",
     1.056
    ],
    "original-adequacy-01-1": [
     "c",
     0.955
    ],
    "original-adequacy-02-0": [
     "c",
     1.22
    ],
    "original-adequacy-02-1": [
     "c",
     1.112
    ],
    "original-adequacy-03-0": [
     "c",
     1.147
    ],
    "original-adequacy-03-1": [
     "c",
     4.354
    ],
    "original-adequacy-04-0": [
     "c",
     1.075
    ],
    "original-adequacy-04-1": [
     "c",
     1.067
    ],
    "original-adequacy-05-0": [
     "c",
     1.331
    ],
    "original-adequacy-05-1": [
     "c",
     1.031
    ],
    "original-adequacy-06-0": [
     "c",
     1.151
    ],
    "original-adequacy-06-1": [
     "c",
     1.151
    ],
    "original-routing-01-0": [
     "c",
     1.054
    ],
    "original-routing-01-1": [
     "c",
     0.728
    ],
    "original-routing-02-0": [
     "c",
     1.465
    ],
    "original-routing-02-1": [
     "c",
     1.066
    ],
    "original-routing-03-0": [
     "c",
     0.959
    ],
    "original-routing-03-1": [
     "c",
     1.111
    ],
    "original-routing-04-0": [
     "c",
     1.663
    ],
    "original-routing-04-1": [
     "c",
     1.63
    ],
    "original-routing-05-0": [
     "c",
     1.401
    ],
    "original-routing-05-1": [
     "c",
     1.746
    ],
    "original-routing-06-0": [
     "c",
     1.068
    ],
    "original-routing-06-1": [
     "c",
     0.859
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     11.69
    ],
    "hard-opus-a-long_policy-04": [
     "c",
     17.357
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     4.775
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     7.235
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     8.269
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     28.935
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     2.851
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     3.62
    ],
    "hard-opus-a-probability-03": [
     "c",
     2.062
    ],
    "hard-opus-a-probability-04": [
     "c",
     5.826
    ],
    "hard-opus-a-probability-07": [
     "c",
     7.072
    ],
    "hard-opus-a-probability-08": [
     "c",
     3.408
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     8.17
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     2.475
    ],
    "hard-opus-a-temporal_numeric-07": [
     "f",
     5.102
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     2.2
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     8.836
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     1.179
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     7.301
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     1.351
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     2.801
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     2.547
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     1.426
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     1.865
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     4.771
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     5.787
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     8.493
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     19.642
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     5.295
    ],
    "hard-opus-b-probability-01": [
     "c",
     6.684
    ],
    "hard-opus-b-probability-02": [
     "c",
     2.157
    ],
    "hard-opus-b-probability-03": [
     "c",
     4.205
    ],
    "hard-opus-b-probability-04": [
     "c",
     1.163
    ],
    "hard-opus-b-probability-06": [
     "f",
     2.957
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     1.746
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     6.742
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     6.176
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     6.873
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     1.912
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     2.127
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     19.932
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     44.2
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     17.078
    ],
    "hard-opus-c-long_policy-05": [
     "c",
     45.389
    ],
    "hard-opus-c-long_policy-08": [
     "c",
     16.276
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     14.795
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     50.008
    ],
    "hard-opus-c-probability-03": [
     "c",
     3.304
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     26.189
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     17.627
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     16.439
    ],
    "hard-opus-c-temporal_numeric-06": [
     "c",
     24.355
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     35.477
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     13.404
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     2.213
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     3.122
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.362
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     2.513
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     10.195
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     1.078
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     2.546
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     2.132
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     2.571
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     1.701
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     2.045
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     9.258
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     4.299
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     3.565
    ],
    "hard-sol-a-trap-02": [
     "c",
     1.166
    ],
    "hard-sol-a-trap-04": [
     "c",
     1.309
    ],
    "hard-sol-a-trap-05": [
     "c",
     6.061
    ],
    "hard-sol-a-trap-06": [
     "c",
     1.058
    ],
    "hard-sol-a-trap-08": [
     "c",
     1.039
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.614
    ],
    "hard-sol-a-trap-13": [
     "c",
     1.857
    ],
    "hard-sol-a-trap-15": [
     "c",
     1.989
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.518
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     1.315
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     1.931
    ],
    "hard-sol-b-judge_hard-08": [
     "c",
     2.035
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     2.27
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     1.52
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     1.418
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     1.656
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     7.553
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     6.792
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     5.298
    ],
    "hard-sol-b-long_policy-06": [
     "f",
     3.748
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.474
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     2.659
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     1.833
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.915
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.363
    ],
    "hard-sol-b-temporal_numeric-01": [
     "c",
     2.53
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     3.068
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     1.9
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     1.724
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     2.25
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     2.479
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     2.494
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     2.751
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     4.277
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     4.177
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     2.157
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     3.519
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.831
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     14.671
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     2.976
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     18.558
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     3.063
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     4.26
    ]
   }
  },
  "needle-3": {
   "display": "Needle 3 (Cactus, 2-bit, local CPU)",
   "partial": true,
   "by_tier": {
    "easy": {
     "c": 34,
     "w": 38,
     "f": 0,
     "n": 0,
     "accuracy": 0.4722
    },
    "standard": {
     "c": 16,
     "w": 80,
     "f": 0,
     "n": 0,
     "accuracy": 0.1667
    },
    "judge": {
     "c": 46,
     "w": 100,
     "f": 0,
     "n": 0,
     "accuracy": 0.3151
    },
    "hard": {
     "c": 17,
     "w": 27,
     "f": 0,
     "n": 176,
     "accuracy": 0.0773
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 4,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.2222
     },
     "fact": {
      "c": 9,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 3,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.1667
     }
    },
    "standard": {
     "policy": {
      "c": 2,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.125
     },
     "intent": {
      "c": 1,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.0625
     },
     "ordinal": {
      "c": 0,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.0
     },
     "extraction": {
      "c": 6,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.375
     },
     "adequacy": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     },
     "routing": {
      "c": 0,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.0
     }
    },
    "judge": {
     "routing": {
      "c": 0,
      "w": 78,
      "f": 0,
      "n": 0,
      "accuracy": 0.0
     },
     "adequacy": {
      "c": 46,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.6765
     }
    },
    "hard": {
     "long_policy": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 26,
      "accuracy": 0.1316
     },
     "probability": {
      "c": 4,
      "w": 5,
      "f": 0,
      "n": 11,
      "accuracy": 0.2
     },
     "temporal_numeric": {
      "c": 3,
      "w": 2,
      "f": 0,
      "n": 25,
      "accuracy": 0.1
     },
     "ambiguous": {
      "c": 3,
      "w": 4,
      "f": 0,
      "n": 7,
      "accuracy": 0.2143
     },
     "multi_hop": {
      "c": 0,
      "w": 5,
      "f": 0,
      "n": 30,
      "accuracy": 0.0
     },
     "tradeoff": {
      "c": 2,
      "w": 4,
      "f": 0,
      "n": 6,
      "accuracy": 0.1667
     },
     "adversarial": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 12,
      "accuracy": 0.0
     },
     "trap": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 16,
      "accuracy": 0.0
     },
     "judge_hard": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 33,
      "accuracy": 0.0
     },
     "routing_hard": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 10,
      "accuracy": 0.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "w",
     3.267
    ],
    "easy-intent-01": [
     "c",
     0.838
    ],
    "easy-intent-02": [
     "w",
     0.102
    ],
    "easy-intent-03": [
     "w",
     0.233
    ],
    "easy-intent-04": [
     "w",
     0.511
    ],
    "easy-intent-05": [
     "w",
     0.877
    ],
    "easy-intent-06": [
     "w",
     1.711
    ],
    "easy-intent-07": [
     "c",
     0.279
    ],
    "easy-intent-08": [
     "w",
     0.223
    ],
    "easy-intent-09": [
     "w",
     0.679
    ],
    "easy-intent-10": [
     "w",
     0.187
    ],
    "easy-intent-11": [
     "c",
     1.386
    ],
    "easy-fact-00": [
     "c",
     24.416
    ],
    "easy-fact-01": [
     "c",
     9.862
    ],
    "easy-fact-02": [
     "c",
     10.314
    ],
    "easy-fact-03": [
     "w",
     2.772
    ],
    "easy-fact-04": [
     "c",
     8.359
    ],
    "easy-fact-05": [
     "w",
     1.466
    ],
    "easy-fact-06": [
     "c",
     3.028
    ],
    "easy-fact-07": [
     "w",
     1.54
    ],
    "easy-fact-08": [
     "c",
     2.841
    ],
    "easy-fact-09": [
     "w",
     0.696
    ],
    "easy-fact-10": [
     "w",
     4.396
    ],
    "easy-fact-11": [
     "w",
     4.328
    ],
    "easy-extraction-00": [
     "c",
     5.566
    ],
    "easy-extraction-01": [
     "c",
     1.794
    ],
    "easy-extraction-02": [
     "c",
     7.94
    ],
    "easy-extraction-03": [
     "c",
     8.522
    ],
    "easy-extraction-04": [
     "c",
     11.28
    ],
    "easy-extraction-05": [
     "c",
     3.283
    ],
    "easy-extraction-06": [
     "c",
     4.619
    ],
    "easy-extraction-07": [
     "c",
     3.591
    ],
    "easy-extraction-08": [
     "c",
     2.444
    ],
    "easy-extraction-09": [
     "c",
     1.954
    ],
    "easy-extraction-10": [
     "c",
     2.397
    ],
    "easy-extraction-11": [
     "c",
     6.262
    ],
    "easy-tool_selection-00": [
     "c",
     4.252
    ],
    "easy-tool_selection-01": [
     "w",
     1.075
    ],
    "easy-tool_selection-02": [
     "w",
     0.259
    ],
    "easy-tool_selection-03": [
     "w",
     0.34
    ],
    "easy-tool_selection-04": [
     "w",
     0.294
    ],
    "easy-tool_selection-05": [
     "w",
     0.753
    ],
    "easy-tool_selection-06": [
     "w",
     0.843
    ],
    "easy-tool_selection-07": [
     "c",
     0.444
    ],
    "easy-tool_selection-08": [
     "w",
     0.125
    ],
    "easy-tool_selection-09": [
     "w",
     0.155
    ],
    "easy-tool_selection-10": [
     "w",
     0.15
    ],
    "easy-tool_selection-11": [
     "w",
     0.098
    ],
    "original-policy-01-0": [
     "w",
     1.524
    ],
    "original-policy-01-1": [
     "w",
     0.316
    ],
    "original-policy-02-0": [
     "c",
     2.091
    ],
    "original-policy-02-1": [
     "w",
     3.045
    ],
    "original-policy-03-0": [
     "w",
     1.314
    ],
    "original-policy-03-1": [
     "w",
     0.534
    ],
    "original-policy-04-0": [
     "w",
     3.312
    ],
    "original-policy-04-1": [
     "w",
     2.739
    ],
    "original-policy-05-0": [
     "w",
     1.475
    ],
    "original-policy-05-1": [
     "w",
     0.504
    ],
    "original-policy-06-0": [
     "w",
     4.191
    ],
    "original-policy-06-1": [
     "w",
     3.157
    ],
    "original-intent-01-0": [
     "w",
     5.617
    ],
    "original-intent-01-1": [
     "c",
     3.205
    ],
    "original-intent-02-0": [
     "w",
     2.224
    ],
    "original-intent-02-1": [
     "w",
     2.492
    ],
    "original-intent-03-0": [
     "w",
     0.887
    ],
    "original-intent-03-1": [
     "w",
     2.895
    ],
    "original-intent-04-0": [
     "w",
     3.252
    ],
    "original-intent-04-1": [
     "w",
     0.867
    ],
    "original-intent-05-0": [
     "w",
     2.236
    ],
    "original-intent-05-1": [
     "w",
     1.782
    ],
    "original-intent-06-0": [
     "w",
     0.78
    ],
    "original-intent-06-1": [
     "w",
     0.849
    ],
    "original-ordinal-01-0": [
     "w",
     5.595
    ],
    "original-ordinal-01-1": [
     "w",
     2.209
    ],
    "original-ordinal-02-0": [
     "w",
     0.431
    ],
    "original-ordinal-02-1": [
     "w",
     0.766
    ],
    "original-ordinal-03-0": [
     "w",
     2.015
    ],
    "original-ordinal-03-1": [
     "w",
     0.605
    ],
    "original-ordinal-04-0": [
     "w",
     0.817
    ],
    "original-ordinal-04-1": [
     "w",
     0.667
    ],
    "original-ordinal-05-0": [
     "w",
     1.072
    ],
    "original-ordinal-05-1": [
     "w",
     2.848
    ],
    "original-ordinal-06-0": [
     "w",
     0.646
    ],
    "original-ordinal-06-1": [
     "w",
     0.238
    ],
    "original-extraction-01-0": [
     "w",
     1.263
    ],
    "original-extraction-01-1": [
     "c",
     0.335
    ],
    "original-extraction-02-0": [
     "w",
     0.464
    ],
    "original-extraction-02-1": [
     "w",
     0.392
    ],
    "original-extraction-03-0": [
     "c",
     0.99
    ],
    "original-extraction-03-1": [
     "w",
     10.069
    ],
    "original-extraction-04-0": [
     "w",
     0.278
    ],
    "original-extraction-04-1": [
     "c",
     0.294
    ],
    "original-extraction-05-0": [
     "w",
     0.38
    ],
    "original-extraction-05-1": [
     "w",
     1.357
    ],
    "original-extraction-06-0": [
     "c",
     1.753
    ],
    "original-extraction-06-1": [
     "c",
     1.374
    ],
    "original-adequacy-01-0": [
     "c",
     0.539
    ],
    "original-adequacy-01-1": [
     "c",
     0.166
    ],
    "original-adequacy-02-0": [
     "w",
     0.397
    ],
    "original-adequacy-02-1": [
     "c",
     0.311
    ],
    "original-adequacy-03-0": [
     "w",
     0.503
    ],
    "original-adequacy-03-1": [
     "w",
     0.394
    ],
    "original-adequacy-04-0": [
     "w",
     0.403
    ],
    "original-adequacy-04-1": [
     "w",
     3.884
    ],
    "original-adequacy-05-0": [
     "w",
     2.233
    ],
    "original-adequacy-05-1": [
     "w",
     0.424
    ],
    "original-adequacy-06-0": [
     "c",
     0.879
    ],
    "original-adequacy-06-1": [
     "c",
     4.181
    ],
    "original-routing-01-0": [
     "w",
     2.561
    ],
    "original-routing-01-1": [
     "w",
     0.366
    ],
    "original-routing-02-0": [
     "w",
     0.406
    ],
    "original-routing-02-1": [
     "w",
     0.248
    ],
    "original-routing-03-0": [
     "w",
     0.844
    ],
    "original-routing-03-1": [
     "w",
     1.365
    ],
    "original-routing-04-0": [
     "w",
     0.163
    ],
    "original-routing-04-1": [
     "w",
     0.112
    ],
    "original-routing-05-0": [
     "w",
     0.119
    ],
    "original-routing-05-1": [
     "w",
     0.155
    ],
    "original-routing-06-0": [
     "w",
     0.158
    ],
    "original-routing-06-1": [
     "w",
     0.689
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     176.354
    ],
    "hard-opus-a-long_policy-04": [
     "c",
     39.191
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     163.576
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     191.285
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     106.505
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     36.513
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     22.924
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     76.92
    ],
    "hard-opus-a-probability-03": [
     "c",
     34.693
    ],
    "hard-opus-a-probability-04": [
     "w",
     23.394
    ],
    "hard-opus-a-probability-07": [
     "w",
     11.592
    ],
    "hard-opus-a-probability-08": [
     "w",
     2.954
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     10.31
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     5.076
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     7.045
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     2.94
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     8.326
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     6.219
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     3.743
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     2.876
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     4.922
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     8.473
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     5.839
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     5.275
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     37.967
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     88.426
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     80.135
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     97.23
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     108.929
    ],
    "hard-opus-b-probability-01": [
     "w",
     44.972
    ],
    "hard-opus-b-probability-02": [
     "w",
     11.921
    ],
    "hard-opus-b-probability-03": [
     "c",
     15.584
    ],
    "hard-opus-b-probability-04": [
     "c",
     7.677
    ],
    "hard-opus-b-probability-06": [
     "c",
     64.796
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     24.863
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     15.617
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     15.223
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     8.768
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     15.644
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     9.263
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     54.415
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     62.996
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     39.015
    ],
    "hard-opus-c-long_policy-05": [
     "c",
     53.642
    ],
    "hard-opus-c-long_policy-08": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-10": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-11": [
     "n",
     null
    ],
    "hard-opus-c-probability-03": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-02": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-03": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-04": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-06": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-08": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-12": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-01": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-06": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-07": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-08": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-09": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-11": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-01": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-02": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-05": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-07": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-08": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-09": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-10": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-12": [
     "n",
     null
    ],
    "hard-sol-a-trap-02": [
     "n",
     null
    ],
    "hard-sol-a-trap-04": [
     "n",
     null
    ],
    "hard-sol-a-trap-05": [
     "n",
     null
    ],
    "hard-sol-a-trap-06": [
     "n",
     null
    ],
    "hard-sol-a-trap-08": [
     "n",
     null
    ],
    "hard-sol-a-trap-10": [
     "n",
     null
    ],
    "hard-sol-a-trap-13": [
     "n",
     null
    ],
    "hard-sol-a-trap-15": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-01": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-02": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-05": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-08": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-10": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-14": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-15": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-18": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-01": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-02": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-05": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-06": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-01": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-02": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-03": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-07": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-09": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-01": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-02": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-03": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-04": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-03": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-05": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-07": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-08": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-09": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-10": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-11": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-13": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-15": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-07": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-09": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-11": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-12": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-13": [
     "n",
     null
    ]
   }
  },
  "qwen3.8-27b": {
   "display": "Qwen3.8 27B (Chutes TEE)",
   "partial": true,
   "by_tier": {
    "easy": {
     "c": 71,
     "w": 0,
     "f": 1,
     "n": 0,
     "accuracy": 0.9861
    },
    "standard": {
     "c": 95,
     "w": 0,
     "f": 1,
     "n": 0,
     "accuracy": 0.9896
    },
    "judge": {
     "c": 121,
     "w": 2,
     "f": 4,
     "n": 19,
     "accuracy": 0.8288
    },
    "hard": {
     "c": 47,
     "w": 1,
     "f": 3,
     "n": 169,
     "accuracy": 0.2136
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 17,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.9444
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "routing": {
      "c": 15,
      "w": 0,
      "f": 1,
      "n": 0,
      "accuracy": 0.9375
     }
    },
    "judge": {
     "routing": {
      "c": 78,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 43,
      "w": 2,
      "f": 4,
      "n": 19,
      "accuracy": 0.6324
     }
    },
    "hard": {
     "long_policy": {
      "c": 12,
      "w": 0,
      "f": 3,
      "n": 23,
      "accuracy": 0.3158
     },
     "probability": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 10,
      "accuracy": 0.5
     },
     "temporal_numeric": {
      "c": 7,
      "w": 1,
      "f": 0,
      "n": 22,
      "accuracy": 0.2333
     },
     "ambiguous": {
      "c": 7,
      "w": 0,
      "f": 0,
      "n": 7,
      "accuracy": 0.5
     },
     "multi_hop": {
      "c": 5,
      "w": 0,
      "f": 0,
      "n": 30,
      "accuracy": 0.1429
     },
     "tradeoff": {
      "c": 6,
      "w": 0,
      "f": 0,
      "n": 6,
      "accuracy": 0.5
     },
     "adversarial": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 12,
      "accuracy": 0.0
     },
     "trap": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 16,
      "accuracy": 0.0
     },
     "judge_hard": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 33,
      "accuracy": 0.0
     },
     "routing_hard": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 10,
      "accuracy": 0.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     5.52
    ],
    "easy-intent-01": [
     "c",
     24.299
    ],
    "easy-intent-02": [
     "c",
     3.617
    ],
    "easy-intent-03": [
     "c",
     2.802
    ],
    "easy-intent-04": [
     "c",
     6.623
    ],
    "easy-intent-05": [
     "c",
     4.2
    ],
    "easy-intent-06": [
     "c",
     3.327
    ],
    "easy-intent-07": [
     "c",
     4.573
    ],
    "easy-intent-08": [
     "c",
     4.017
    ],
    "easy-intent-09": [
     "c",
     4.565
    ],
    "easy-intent-10": [
     "c",
     14.937
    ],
    "easy-intent-11": [
     "c",
     4.496
    ],
    "easy-fact-00": [
     "c",
     3.172
    ],
    "easy-fact-01": [
     "c",
     2.479
    ],
    "easy-fact-02": [
     "c",
     3.54
    ],
    "easy-fact-03": [
     "c",
     3.323
    ],
    "easy-fact-04": [
     "c",
     11.686
    ],
    "easy-fact-05": [
     "c",
     4.264
    ],
    "easy-fact-06": [
     "c",
     2.963
    ],
    "easy-fact-07": [
     "c",
     2.901
    ],
    "easy-fact-08": [
     "c",
     8.639
    ],
    "easy-fact-09": [
     "c",
     3.109
    ],
    "easy-fact-10": [
     "c",
     2.663
    ],
    "easy-fact-11": [
     "c",
     2.43
    ],
    "easy-extraction-00": [
     "c",
     2.988
    ],
    "easy-extraction-01": [
     "c",
     3.815
    ],
    "easy-extraction-02": [
     "c",
     3.248
    ],
    "easy-extraction-03": [
     "c",
     3.276
    ],
    "easy-extraction-04": [
     "c",
     3.138
    ],
    "easy-extraction-05": [
     "c",
     2.66
    ],
    "easy-extraction-06": [
     "c",
     2.99
    ],
    "easy-extraction-07": [
     "c",
     6.095
    ],
    "easy-extraction-08": [
     "c",
     4.165
    ],
    "easy-extraction-09": [
     "c",
     2.579
    ],
    "easy-extraction-10": [
     "c",
     2.748
    ],
    "easy-extraction-11": [
     "c",
     3.034
    ],
    "easy-tool_selection-00": [
     "c",
     5.2
    ],
    "easy-tool_selection-01": [
     "c",
     4.612
    ],
    "easy-tool_selection-02": [
     "c",
     4.59
    ],
    "easy-tool_selection-03": [
     "c",
     4.538
    ],
    "easy-tool_selection-04": [
     "c",
     3.088
    ],
    "easy-tool_selection-05": [
     "c",
     11.72
    ],
    "easy-tool_selection-06": [
     "c",
     4.877
    ],
    "easy-tool_selection-07": [
     "c",
     4.536
    ],
    "easy-tool_selection-08": [
     "c",
     3.841
    ],
    "easy-tool_selection-09": [
     "c",
     3.673
    ],
    "easy-tool_selection-10": [
     "c",
     4.672
    ],
    "easy-tool_selection-11": [
     "c",
     3.349
    ],
    "original-policy-01-0": [
     "c",
     5.253
    ],
    "original-policy-01-1": [
     "c",
     3.685
    ],
    "original-policy-02-0": [
     "c",
     5.001
    ],
    "original-policy-02-1": [
     "c",
     3.568
    ],
    "original-policy-03-0": [
     "c",
     5.546
    ],
    "original-policy-03-1": [
     "c",
     5.553
    ],
    "original-policy-04-0": [
     "c",
     5.832
    ],
    "original-policy-04-1": [
     "c",
     5.772
    ],
    "original-policy-05-0": [
     "c",
     4.813
    ],
    "original-policy-05-1": [
     "c",
     4.226
    ],
    "original-policy-06-0": [
     "c",
     6.211
    ],
    "original-policy-06-1": [
     "c",
     6.14
    ],
    "original-intent-01-0": [
     "c",
     5.006
    ],
    "original-intent-01-1": [
     "c",
     9.328
    ],
    "original-intent-02-0": [
     "c",
     5.265
    ],
    "original-intent-02-1": [
     "c",
     4.554
    ],
    "original-intent-03-0": [
     "c",
     5.89
    ],
    "original-intent-03-1": [
     "c",
     4.565
    ],
    "original-intent-04-0": [
     "c",
     4.968
    ],
    "original-intent-04-1": [
     "c",
     6.919
    ],
    "original-intent-05-0": [
     "c",
     6.194
    ],
    "original-intent-05-1": [
     "c",
     8.337
    ],
    "original-intent-06-0": [
     "c",
     6.625
    ],
    "original-intent-06-1": [
     "c",
     4.987
    ],
    "original-ordinal-01-0": [
     "c",
     3.694
    ],
    "original-ordinal-01-1": [
     "c",
     3.7
    ],
    "original-ordinal-02-0": [
     "c",
     6.794
    ],
    "original-ordinal-02-1": [
     "c",
     4.899
    ],
    "original-ordinal-03-0": [
     "c",
     4.448
    ],
    "original-ordinal-03-1": [
     "c",
     4.292
    ],
    "original-ordinal-04-0": [
     "c",
     4.824
    ],
    "original-ordinal-04-1": [
     "c",
     4.397
    ],
    "original-ordinal-05-0": [
     "c",
     4.376
    ],
    "original-ordinal-05-1": [
     "c",
     4.167
    ],
    "original-ordinal-06-0": [
     "c",
     5.744
    ],
    "original-ordinal-06-1": [
     "c",
     5.365
    ],
    "original-extraction-01-0": [
     "c",
     3.698
    ],
    "original-extraction-01-1": [
     "c",
     3.605
    ],
    "original-extraction-02-0": [
     "c",
     5.145
    ],
    "original-extraction-02-1": [
     "c",
     3.775
    ],
    "original-extraction-03-0": [
     "c",
     6.101
    ],
    "original-extraction-03-1": [
     "c",
     4.345
    ],
    "original-extraction-04-0": [
     "c",
     3.768
    ],
    "original-extraction-04-1": [
     "c",
     3.493
    ],
    "original-extraction-05-0": [
     "c",
     3.16
    ],
    "original-extraction-05-1": [
     "c",
     3.457
    ],
    "original-extraction-06-0": [
     "c",
     4.027
    ],
    "original-extraction-06-1": [
     "c",
     4.785
    ],
    "original-adequacy-01-0": [
     "c",
     3.223
    ],
    "original-adequacy-01-1": [
     "c",
     10.394
    ],
    "original-adequacy-02-0": [
     "c",
     3.505
    ],
    "original-adequacy-02-1": [
     "c",
     3.418
    ],
    "original-adequacy-03-0": [
     "c",
     3.629
    ],
    "original-adequacy-03-1": [
     "c",
     68.971
    ],
    "original-adequacy-04-0": [
     "c",
     4.893
    ],
    "original-adequacy-04-1": [
     "c",
     3.515
    ],
    "original-adequacy-05-0": [
     "c",
     3.303
    ],
    "original-adequacy-05-1": [
     "c",
     3.57
    ],
    "original-adequacy-06-0": [
     "c",
     3.776
    ],
    "original-adequacy-06-1": [
     "c",
     8.712
    ],
    "original-routing-01-0": [
     "c",
     6.785
    ],
    "original-routing-01-1": [
     "c",
     8.035
    ],
    "original-routing-02-0": [
     "c",
     6.664
    ],
    "original-routing-02-1": [
     "c",
     9.85
    ],
    "original-routing-03-0": [
     "c",
     4.197
    ],
    "original-routing-03-1": [
     "c",
     3.259
    ],
    "original-routing-04-0": [
     "c",
     4.972
    ],
    "original-routing-04-1": [
     "c",
     6.387
    ],
    "original-routing-05-0": [
     "c",
     4.794
    ],
    "original-routing-05-1": [
     "c",
     4.058
    ],
    "original-routing-06-0": [
     "f",
     0.348
    ],
    "original-routing-06-1": [
     "c",
     4.984
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     31.78
    ],
    "hard-opus-a-long_policy-04": [
     "c",
     75.468
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     54.62
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     40.01
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     35.83
    ],
    "hard-opus-a-long_policy-13": [
     "f",
     120.114
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     15.646
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     17.882
    ],
    "hard-opus-a-probability-03": [
     "c",
     8.159
    ],
    "hard-opus-a-probability-04": [
     "c",
     21.577
    ],
    "hard-opus-a-probability-07": [
     "c",
     22.874
    ],
    "hard-opus-a-probability-08": [
     "c",
     9.454
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     22.817
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     13.059
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     26.574
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     11.03
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     30.957
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     8.725
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     19.95
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     5.773
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     13.701
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     9.489
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     11.612
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     7.282
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     20.77
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     16.93
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     17.386
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     44.487
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     18.098
    ],
    "hard-opus-b-probability-01": [
     "c",
     24.17
    ],
    "hard-opus-b-probability-02": [
     "c",
     6.402
    ],
    "hard-opus-b-probability-03": [
     "c",
     20.173
    ],
    "hard-opus-b-probability-04": [
     "c",
     5.632
    ],
    "hard-opus-b-probability-06": [
     "c",
     14.271
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     10.166
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     21.825
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     40.068
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     15.507
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     9.819
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     13.447
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     44.523
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     88.004
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     91.669
    ],
    "hard-opus-c-long_policy-05": [
     "f",
     120.146
    ],
    "hard-opus-c-long_policy-08": [
     "c",
     60.353
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     29.151
    ],
    "hard-opus-c-long_policy-11": [
     "f",
     120.112
    ],
    "hard-opus-c-probability-03": [
     "c",
     72.618
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     68.851
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     39.303
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     54.556
    ],
    "hard-opus-c-temporal_numeric-06": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-08": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-12": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-01": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-06": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-07": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-08": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-09": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-11": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-01": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-02": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-05": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-07": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-08": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-09": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-10": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-12": [
     "n",
     null
    ],
    "hard-sol-a-trap-02": [
     "n",
     null
    ],
    "hard-sol-a-trap-04": [
     "n",
     null
    ],
    "hard-sol-a-trap-05": [
     "n",
     null
    ],
    "hard-sol-a-trap-06": [
     "n",
     null
    ],
    "hard-sol-a-trap-08": [
     "n",
     null
    ],
    "hard-sol-a-trap-10": [
     "n",
     null
    ],
    "hard-sol-a-trap-13": [
     "n",
     null
    ],
    "hard-sol-a-trap-15": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-01": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-02": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-05": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-08": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-10": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-14": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-15": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-18": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-01": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-02": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-05": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-06": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-01": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-02": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-03": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-07": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-09": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-01": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-02": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-03": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-04": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-03": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-05": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-07": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-08": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-09": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-10": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-11": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-13": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-15": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-07": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-09": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-11": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-12": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-13": [
     "n",
     null
    ]
   }
  },
  "needle-3-tools": {
   "display": "Needle 3, options as tools (post-hoc adapter mode)",
   "partial": true,
   "by_tier": {
    "easy": {
     "c": 48,
     "w": 24,
     "f": 0,
     "n": 0,
     "accuracy": 0.6667
    },
    "standard": {
     "c": 30,
     "w": 66,
     "f": 0,
     "n": 0,
     "accuracy": 0.3125
    },
    "judge": {
     "c": 50,
     "w": 96,
     "f": 0,
     "n": 0,
     "accuracy": 0.3425
    },
    "hard": {
     "c": 0,
     "w": 0,
     "f": 0,
     "n": 220,
     "accuracy": 0.0
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 14,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.7778
     },
     "fact": {
      "c": 9,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "extraction": {
      "c": 8,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.4444
     },
     "tool_selection": {
      "c": 17,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9444
     }
    },
    "standard": {
     "policy": {
      "c": 1,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.0625
     },
     "intent": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "ordinal": {
      "c": 1,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.0625
     },
     "extraction": {
      "c": 4,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "adequacy": {
      "c": 6,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.375
     },
     "routing": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     }
    },
    "judge": {
     "routing": {
      "c": 4,
      "w": 74,
      "f": 0,
      "n": 0,
      "accuracy": 0.0513
     },
     "adequacy": {
      "c": 46,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.6765
     }
    },
    "hard": {
     "long_policy": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 38,
      "accuracy": 0.0
     },
     "probability": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 20,
      "accuracy": 0.0
     },
     "temporal_numeric": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 30,
      "accuracy": 0.0
     },
     "ambiguous": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 14,
      "accuracy": 0.0
     },
     "multi_hop": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 35,
      "accuracy": 0.0
     },
     "tradeoff": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 12,
      "accuracy": 0.0
     },
     "adversarial": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 12,
      "accuracy": 0.0
     },
     "trap": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 16,
      "accuracy": 0.0
     },
     "judge_hard": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 33,
      "accuracy": 0.0
     },
     "routing_hard": {
      "c": 0,
      "w": 0,
      "f": 0,
      "n": 10,
      "accuracy": 0.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.403
    ],
    "easy-intent-01": [
     "c",
     0.175
    ],
    "easy-intent-02": [
     "w",
     0.142
    ],
    "easy-intent-03": [
     "c",
     0.324
    ],
    "easy-intent-04": [
     "w",
     0.273
    ],
    "easy-intent-05": [
     "c",
     0.275
    ],
    "easy-intent-06": [
     "c",
     0.528
    ],
    "easy-intent-07": [
     "c",
     0.14
    ],
    "easy-intent-08": [
     "c",
     0.164
    ],
    "easy-intent-09": [
     "c",
     0.223
    ],
    "easy-intent-10": [
     "c",
     0.437
    ],
    "easy-intent-11": [
     "c",
     0.186
    ],
    "easy-fact-00": [
     "c",
     0.868
    ],
    "easy-fact-01": [
     "c",
     0.571
    ],
    "easy-fact-02": [
     "c",
     1.551
    ],
    "easy-fact-03": [
     "c",
     0.44
    ],
    "easy-fact-04": [
     "w",
     0.342
    ],
    "easy-fact-05": [
     "w",
     0.158
    ],
    "easy-fact-06": [
     "c",
     0.775
    ],
    "easy-fact-07": [
     "w",
     0.758
    ],
    "easy-fact-08": [
     "c",
     0.431
    ],
    "easy-fact-09": [
     "w",
     0.142
    ],
    "easy-fact-10": [
     "w",
     3.558
    ],
    "easy-fact-11": [
     "w",
     2.167
    ],
    "easy-extraction-00": [
     "w",
     1.797
    ],
    "easy-extraction-01": [
     "w",
     0.292
    ],
    "easy-extraction-02": [
     "w",
     0.232
    ],
    "easy-extraction-03": [
     "c",
     0.64
    ],
    "easy-extraction-04": [
     "w",
     0.37
    ],
    "easy-extraction-05": [
     "c",
     0.198
    ],
    "easy-extraction-06": [
     "c",
     0.254
    ],
    "easy-extraction-07": [
     "w",
     0.806
    ],
    "easy-extraction-08": [
     "c",
     1.281
    ],
    "easy-extraction-09": [
     "w",
     0.484
    ],
    "easy-extraction-10": [
     "w",
     0.158
    ],
    "easy-extraction-11": [
     "c",
     0.304
    ],
    "easy-tool_selection-00": [
     "c",
     0.385
    ],
    "easy-tool_selection-01": [
     "c",
     0.315
    ],
    "easy-tool_selection-02": [
     "c",
     0.312
    ],
    "easy-tool_selection-03": [
     "c",
     0.298
    ],
    "easy-tool_selection-04": [
     "c",
     0.155
    ],
    "easy-tool_selection-05": [
     "c",
     0.342
    ],
    "easy-tool_selection-06": [
     "c",
     0.748
    ],
    "easy-tool_selection-07": [
     "c",
     0.211
    ],
    "easy-tool_selection-08": [
     "w",
     0.198
    ],
    "easy-tool_selection-09": [
     "c",
     0.216
    ],
    "easy-tool_selection-10": [
     "c",
     0.182
    ],
    "easy-tool_selection-11": [
     "c",
     0.188
    ],
    "original-policy-01-0": [
     "w",
     8.458
    ],
    "original-policy-01-1": [
     "w",
     2.385
    ],
    "original-policy-02-0": [
     "w",
     0.533
    ],
    "original-policy-02-1": [
     "w",
     0.526
    ],
    "original-policy-03-0": [
     "w",
     0.473
    ],
    "original-policy-03-1": [
     "w",
     0.723
    ],
    "original-policy-04-0": [
     "w",
     0.666
    ],
    "original-policy-04-1": [
     "w",
     0.359
    ],
    "original-policy-05-0": [
     "w",
     0.32
    ],
    "original-policy-05-1": [
     "w",
     2.957
    ],
    "original-policy-06-0": [
     "w",
     0.897
    ],
    "original-policy-06-1": [
     "w",
     0.301
    ],
    "original-intent-01-0": [
     "c",
     2.103
    ],
    "original-intent-01-1": [
     "w",
     0.952
    ],
    "original-intent-02-0": [
     "c",
     0.502
    ],
    "original-intent-02-1": [
     "c",
     0.522
    ],
    "original-intent-03-0": [
     "w",
     0.505
    ],
    "original-intent-03-1": [
     "c",
     0.956
    ],
    "original-intent-04-0": [
     "c",
     2.593
    ],
    "original-intent-04-1": [
     "c",
     3.391
    ],
    "original-intent-05-0": [
     "w",
     0.86
    ],
    "original-intent-05-1": [
     "w",
     0.256
    ],
    "original-intent-06-0": [
     "c",
     0.299
    ],
    "original-intent-06-1": [
     "c",
     0.278
    ],
    "original-ordinal-01-0": [
     "w",
     1.27
    ],
    "original-ordinal-01-1": [
     "w",
     0.668
    ],
    "original-ordinal-02-0": [
     "w",
     0.325
    ],
    "original-ordinal-02-1": [
     "w",
     0.188
    ],
    "original-ordinal-03-0": [
     "w",
     0.714
    ],
    "original-ordinal-03-1": [
     "w",
     0.356
    ],
    "original-ordinal-04-0": [
     "w",
     0.736
    ],
    "original-ordinal-04-1": [
     "w",
     0.373
    ],
    "original-ordinal-05-0": [
     "w",
     2.857
    ],
    "original-ordinal-05-1": [
     "w",
     3.003
    ],
    "original-ordinal-06-0": [
     "w",
     1.529
    ],
    "original-ordinal-06-1": [
     "w",
     3.79
    ],
    "original-extraction-01-0": [
     "w",
     8.919
    ],
    "original-extraction-01-1": [
     "w",
     1.741
    ],
    "original-extraction-02-0": [
     "w",
     0.656
    ],
    "original-extraction-02-1": [
     "w",
     0.85
    ],
    "original-extraction-03-0": [
     "w",
     1.512
    ],
    "original-extraction-03-1": [
     "w",
     7.192
    ],
    "original-extraction-04-0": [
     "w",
     1.819
    ],
    "original-extraction-04-1": [
     "c",
     0.966
    ],
    "original-extraction-05-0": [
     "w",
     0.476
    ],
    "original-extraction-05-1": [
     "w",
     3.179
    ],
    "original-extraction-06-0": [
     "c",
     6.09
    ],
    "original-extraction-06-1": [
     "w",
     2.29
    ],
    "original-adequacy-01-0": [
     "c",
     2.364
    ],
    "original-adequacy-01-1": [
     "c",
     1.949
    ],
    "original-adequacy-02-0": [
     "w",
     5.401
    ],
    "original-adequacy-02-1": [
     "c",
     3.21
    ],
    "original-adequacy-03-0": [
     "w",
     5.041
    ],
    "original-adequacy-03-1": [
     "w",
     4.881
    ],
    "original-adequacy-04-0": [
     "w",
     3.625
    ],
    "original-adequacy-04-1": [
     "w",
     7.42
    ],
    "original-adequacy-05-0": [
     "w",
     1.234
    ],
    "original-adequacy-05-1": [
     "w",
     0.866
    ],
    "original-adequacy-06-0": [
     "c",
     9.881
    ],
    "original-adequacy-06-1": [
     "c",
     10.788
    ],
    "original-routing-01-0": [
     "c",
     7.21
    ],
    "original-routing-01-1": [
     "w",
     3.423
    ],
    "original-routing-02-0": [
     "w",
     1.746
    ],
    "original-routing-02-1": [
     "w",
     7.817
    ],
    "original-routing-03-0": [
     "w",
     6.605
    ],
    "original-routing-03-1": [
     "w",
     5.778
    ],
    "original-routing-04-0": [
     "c",
     5.53
    ],
    "original-routing-04-1": [
     "c",
     4.505
    ],
    "original-routing-05-0": [
     "w",
     3.235
    ],
    "original-routing-05-1": [
     "w",
     0.859
    ],
    "original-routing-06-0": [
     "c",
     3.941
    ],
    "original-routing-06-1": [
     "w",
     1.948
    ],
    "hard-opus-a-long_policy-01": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-04": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-08": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-09": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-11": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-13": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-17": [
     "n",
     null
    ],
    "hard-opus-a-long_policy-19": [
     "n",
     null
    ],
    "hard-opus-a-probability-03": [
     "n",
     null
    ],
    "hard-opus-a-probability-04": [
     "n",
     null
    ],
    "hard-opus-a-probability-07": [
     "n",
     null
    ],
    "hard-opus-a-probability-08": [
     "n",
     null
    ],
    "hard-opus-a-temporal_numeric-03": [
     "n",
     null
    ],
    "hard-opus-a-temporal_numeric-06": [
     "n",
     null
    ],
    "hard-opus-a-temporal_numeric-07": [
     "n",
     null
    ],
    "hard-opus-a-temporal_numeric-09": [
     "n",
     null
    ],
    "hard-opus-a-temporal_numeric-12": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-02": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-03": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-07": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-09": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-10": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-11": [
     "n",
     null
    ],
    "hard-opus-b-ambiguous-13": [
     "n",
     null
    ],
    "hard-opus-b-multi_hop-03": [
     "n",
     null
    ],
    "hard-opus-b-multi_hop-04": [
     "n",
     null
    ],
    "hard-opus-b-multi_hop-05": [
     "n",
     null
    ],
    "hard-opus-b-multi_hop-07": [
     "n",
     null
    ],
    "hard-opus-b-multi_hop-08": [
     "n",
     null
    ],
    "hard-opus-b-probability-01": [
     "n",
     null
    ],
    "hard-opus-b-probability-02": [
     "n",
     null
    ],
    "hard-opus-b-probability-03": [
     "n",
     null
    ],
    "hard-opus-b-probability-04": [
     "n",
     null
    ],
    "hard-opus-b-probability-06": [
     "n",
     null
    ],
    "hard-opus-b-tradeoff-01": [
     "n",
     null
    ],
    "hard-opus-b-tradeoff-03": [
     "n",
     null
    ],
    "hard-opus-b-tradeoff-06": [
     "n",
     null
    ],
    "hard-opus-b-tradeoff-07": [
     "n",
     null
    ],
    "hard-opus-b-tradeoff-08": [
     "n",
     null
    ],
    "hard-opus-b-tradeoff-12": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-02": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-03": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-04": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-05": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-08": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-10": [
     "n",
     null
    ],
    "hard-opus-c-long_policy-11": [
     "n",
     null
    ],
    "hard-opus-c-probability-03": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-02": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-03": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-04": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-06": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-08": [
     "n",
     null
    ],
    "hard-opus-c-temporal_numeric-12": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-01": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-06": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-07": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-08": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-09": [
     "n",
     null
    ],
    "hard-sol-a-adversarial-11": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-01": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-02": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-05": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-07": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-08": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-09": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-10": [
     "n",
     null
    ],
    "hard-sol-a-multi_hop-12": [
     "n",
     null
    ],
    "hard-sol-a-trap-02": [
     "n",
     null
    ],
    "hard-sol-a-trap-04": [
     "n",
     null
    ],
    "hard-sol-a-trap-05": [
     "n",
     null
    ],
    "hard-sol-a-trap-06": [
     "n",
     null
    ],
    "hard-sol-a-trap-08": [
     "n",
     null
    ],
    "hard-sol-a-trap-10": [
     "n",
     null
    ],
    "hard-sol-a-trap-13": [
     "n",
     null
    ],
    "hard-sol-a-trap-15": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-01": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-02": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-05": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-08": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-10": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-14": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-15": [
     "n",
     null
    ],
    "hard-sol-b-judge_hard-18": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-01": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-02": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-05": [
     "n",
     null
    ],
    "hard-sol-b-long_policy-06": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-01": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-02": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-03": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-07": [
     "n",
     null
    ],
    "hard-sol-b-routing_hard-09": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-01": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-02": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-03": [
     "n",
     null
    ],
    "hard-sol-b-temporal_numeric-04": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-03": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-05": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-07": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-08": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-09": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-10": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-11": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-13": [
     "n",
     null
    ],
    "hard-sol-c-judge_hard-15": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-07": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-09": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-11": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-12": [
     "n",
     null
    ],
    "hard-sol-c-multi_hop-13": [
     "n",
     null
    ]
   }
  },
  "open-alternative-jev": {
   "display": "open-alternative-jev (Qwen3.5-4B, IkerMoel)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 81,
     "w": 15,
     "f": 0,
     "n": 0,
     "accuracy": 0.8438
    },
    "judge": {
     "c": 109,
     "w": 37,
     "f": 0,
     "n": 0,
     "accuracy": 0.7466
    },
    "hard": {
     "c": 125,
     "w": 95,
     "f": 0,
     "n": 0,
     "accuracy": 0.5682
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "intent": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "routing": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     }
    },
    "judge": {
     "routing": {
      "c": 60,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.7692
     },
     "adequacy": {
      "c": 49,
      "w": 19,
      "f": 0,
      "n": 0,
      "accuracy": 0.7206
     }
    },
    "hard": {
     "long_policy": {
      "c": 17,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.4474
     },
     "probability": {
      "c": 7,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.35
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 9,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6429
     },
     "multi_hop": {
      "c": 19,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.5429
     },
     "tradeoff": {
      "c": 7,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.5833
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 23,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.697
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.498
    ],
    "easy-intent-01": [
     "c",
     0.197
    ],
    "easy-intent-02": [
     "c",
     0.195
    ],
    "easy-intent-03": [
     "c",
     0.187
    ],
    "easy-intent-04": [
     "c",
     0.188
    ],
    "easy-intent-05": [
     "c",
     0.201
    ],
    "easy-intent-06": [
     "c",
     0.198
    ],
    "easy-intent-07": [
     "c",
     0.197
    ],
    "easy-intent-08": [
     "c",
     0.192
    ],
    "easy-intent-09": [
     "c",
     0.19
    ],
    "easy-intent-10": [
     "c",
     0.188
    ],
    "easy-intent-11": [
     "c",
     0.193
    ],
    "easy-fact-00": [
     "c",
     0.192
    ],
    "easy-fact-01": [
     "c",
     0.185
    ],
    "easy-fact-02": [
     "c",
     0.185
    ],
    "easy-fact-03": [
     "c",
     0.194
    ],
    "easy-fact-04": [
     "c",
     0.189
    ],
    "easy-fact-05": [
     "c",
     0.194
    ],
    "easy-fact-06": [
     "c",
     0.192
    ],
    "easy-fact-07": [
     "c",
     0.19
    ],
    "easy-fact-08": [
     "c",
     0.187
    ],
    "easy-fact-09": [
     "c",
     0.191
    ],
    "easy-fact-10": [
     "c",
     0.192
    ],
    "easy-fact-11": [
     "c",
     0.187
    ],
    "easy-extraction-00": [
     "c",
     0.19
    ],
    "easy-extraction-01": [
     "c",
     0.186
    ],
    "easy-extraction-02": [
     "c",
     0.188
    ],
    "easy-extraction-03": [
     "c",
     0.189
    ],
    "easy-extraction-04": [
     "c",
     0.19
    ],
    "easy-extraction-05": [
     "c",
     0.184
    ],
    "easy-extraction-06": [
     "c",
     0.186
    ],
    "easy-extraction-07": [
     "c",
     0.19
    ],
    "easy-extraction-08": [
     "c",
     0.184
    ],
    "easy-extraction-09": [
     "c",
     0.185
    ],
    "easy-extraction-10": [
     "c",
     0.183
    ],
    "easy-extraction-11": [
     "c",
     0.19
    ],
    "easy-tool_selection-00": [
     "c",
     0.195
    ],
    "easy-tool_selection-01": [
     "c",
     0.197
    ],
    "easy-tool_selection-02": [
     "c",
     0.198
    ],
    "easy-tool_selection-03": [
     "c",
     0.195
    ],
    "easy-tool_selection-04": [
     "c",
     0.199
    ],
    "easy-tool_selection-05": [
     "c",
     0.19
    ],
    "easy-tool_selection-06": [
     "c",
     0.194
    ],
    "easy-tool_selection-07": [
     "c",
     0.197
    ],
    "easy-tool_selection-08": [
     "c",
     0.197
    ],
    "easy-tool_selection-09": [
     "c",
     0.19
    ],
    "easy-tool_selection-10": [
     "c",
     0.196
    ],
    "easy-tool_selection-11": [
     "c",
     0.2
    ],
    "original-policy-01-0": [
     "c",
     0.195
    ],
    "original-policy-01-1": [
     "c",
     0.189
    ],
    "original-policy-02-0": [
     "c",
     0.191
    ],
    "original-policy-02-1": [
     "w",
     0.185
    ],
    "original-policy-03-0": [
     "c",
     0.185
    ],
    "original-policy-03-1": [
     "w",
     0.192
    ],
    "original-policy-04-0": [
     "c",
     0.188
    ],
    "original-policy-04-1": [
     "c",
     0.19
    ],
    "original-policy-05-0": [
     "c",
     0.194
    ],
    "original-policy-05-1": [
     "c",
     0.189
    ],
    "original-policy-06-0": [
     "c",
     0.185
    ],
    "original-policy-06-1": [
     "c",
     0.192
    ],
    "original-intent-01-0": [
     "c",
     0.197
    ],
    "original-intent-01-1": [
     "c",
     0.194
    ],
    "original-intent-02-0": [
     "w",
     0.198
    ],
    "original-intent-02-1": [
     "c",
     0.199
    ],
    "original-intent-03-0": [
     "c",
     0.19
    ],
    "original-intent-03-1": [
     "c",
     0.2
    ],
    "original-intent-04-0": [
     "c",
     0.197
    ],
    "original-intent-04-1": [
     "c",
     0.193
    ],
    "original-intent-05-0": [
     "w",
     0.187
    ],
    "original-intent-05-1": [
     "w",
     0.194
    ],
    "original-intent-06-0": [
     "c",
     0.194
    ],
    "original-intent-06-1": [
     "c",
     0.199
    ],
    "original-ordinal-01-0": [
     "c",
     0.196
    ],
    "original-ordinal-01-1": [
     "c",
     0.214
    ],
    "original-ordinal-02-0": [
     "c",
     0.211
    ],
    "original-ordinal-02-1": [
     "c",
     0.207
    ],
    "original-ordinal-03-0": [
     "c",
     0.211
    ],
    "original-ordinal-03-1": [
     "c",
     0.208
    ],
    "original-ordinal-04-0": [
     "c",
     0.195
    ],
    "original-ordinal-04-1": [
     "c",
     0.191
    ],
    "original-ordinal-05-0": [
     "c",
     0.193
    ],
    "original-ordinal-05-1": [
     "c",
     0.193
    ],
    "original-ordinal-06-0": [
     "c",
     0.195
    ],
    "original-ordinal-06-1": [
     "c",
     0.188
    ],
    "original-extraction-01-0": [
     "c",
     0.186
    ],
    "original-extraction-01-1": [
     "c",
     0.187
    ],
    "original-extraction-02-0": [
     "c",
     0.181
    ],
    "original-extraction-02-1": [
     "c",
     0.187
    ],
    "original-extraction-03-0": [
     "c",
     0.187
    ],
    "original-extraction-03-1": [
     "c",
     0.184
    ],
    "original-extraction-04-0": [
     "c",
     0.19
    ],
    "original-extraction-04-1": [
     "c",
     0.184
    ],
    "original-extraction-05-0": [
     "c",
     0.196
    ],
    "original-extraction-05-1": [
     "c",
     0.189
    ],
    "original-extraction-06-0": [
     "c",
     0.192
    ],
    "original-extraction-06-1": [
     "c",
     0.186
    ],
    "original-adequacy-01-0": [
     "c",
     0.186
    ],
    "original-adequacy-01-1": [
     "c",
     0.186
    ],
    "original-adequacy-02-0": [
     "w",
     0.192
    ],
    "original-adequacy-02-1": [
     "c",
     0.181
    ],
    "original-adequacy-03-0": [
     "w",
     0.191
    ],
    "original-adequacy-03-1": [
     "w",
     0.185
    ],
    "original-adequacy-04-0": [
     "c",
     0.185
    ],
    "original-adequacy-04-1": [
     "c",
     0.188
    ],
    "original-adequacy-05-0": [
     "w",
     0.192
    ],
    "original-adequacy-05-1": [
     "w",
     0.187
    ],
    "original-adequacy-06-0": [
     "c",
     0.196
    ],
    "original-adequacy-06-1": [
     "c",
     0.186
    ],
    "original-routing-01-0": [
     "c",
     0.198
    ],
    "original-routing-01-1": [
     "c",
     0.196
    ],
    "original-routing-02-0": [
     "c",
     0.196
    ],
    "original-routing-02-1": [
     "c",
     0.19
    ],
    "original-routing-03-0": [
     "c",
     0.189
    ],
    "original-routing-03-1": [
     "c",
     0.194
    ],
    "original-routing-04-0": [
     "c",
     0.188
    ],
    "original-routing-04-1": [
     "w",
     0.199
    ],
    "original-routing-05-0": [
     "c",
     0.196
    ],
    "original-routing-05-1": [
     "w",
     0.194
    ],
    "original-routing-06-0": [
     "c",
     0.194
    ],
    "original-routing-06-1": [
     "c",
     0.189
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     1.992
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.474
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.44
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.45
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.436
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.437
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.432
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.434
    ],
    "hard-opus-a-probability-03": [
     "w",
     0.247
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.293
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.303
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.216
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.231
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.237
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     0.345
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.23
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.234
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.236
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.225
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.228
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.274
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.246
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.244
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.252
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.528
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.454
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.475
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.47
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.428
    ],
    "hard-opus-b-probability-01": [
     "w",
     0.379
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.228
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.299
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.236
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.29
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.269
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.263
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.264
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.262
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.241
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.248
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.676
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.677
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.621
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.625
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.626
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     0.676
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.63
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.233
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.273
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.408
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.267
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.294
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.287
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.272
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.212
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.212
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.212
    ],
    "hard-sol-a-adversarial-08": [
     "w",
     0.203
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.199
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.217
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.431
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.435
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.437
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.438
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.43
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.436
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.439
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.436
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.203
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.203
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.22
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.219
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.205
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.195
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.195
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.206
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.209
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.198
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.199
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.202
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.202
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.207
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.205
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.194
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.549
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     0.524
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.535
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.531
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.209
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.215
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.209
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.212
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.206
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.209
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.216
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.203
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.209
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.225
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     0.217
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.213
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.218
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.215
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.214
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.215
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.217
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.205
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.461
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.482
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.479
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.238
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.473
    ]
   }
  },
  "classifier-dev-fast": {
   "display": "classifier.dev (fast tier)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 95,
     "w": 1,
     "f": 0,
     "n": 0,
     "accuracy": 0.9896
    },
    "judge": {
     "c": 142,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9726
    },
    "hard": {
     "c": 155,
     "w": 65,
     "f": 0,
     "n": 0,
     "accuracy": 0.7045
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 78,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 64,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.9412
     }
    },
    "hard": {
     "long_policy": {
      "c": 20,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.5263
     },
     "probability": {
      "c": 14,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.7
     },
     "temporal_numeric": {
      "c": 7,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.2333
     },
     "ambiguous": {
      "c": 11,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.7857
     },
     "multi_hop": {
      "c": 28,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8
     },
     "tradeoff": {
      "c": 11,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9167
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 26,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.7879
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.399
    ],
    "easy-intent-01": [
     "c",
     0.356
    ],
    "easy-intent-02": [
     "c",
     0.519
    ],
    "easy-intent-03": [
     "c",
     0.364
    ],
    "easy-intent-04": [
     "c",
     0.431
    ],
    "easy-intent-05": [
     "c",
     0.378
    ],
    "easy-intent-06": [
     "c",
     0.405
    ],
    "easy-intent-07": [
     "c",
     0.433
    ],
    "easy-intent-08": [
     "c",
     0.372
    ],
    "easy-intent-09": [
     "c",
     0.416
    ],
    "easy-intent-10": [
     "c",
     0.413
    ],
    "easy-intent-11": [
     "c",
     0.493
    ],
    "easy-fact-00": [
     "c",
     0.454
    ],
    "easy-fact-01": [
     "c",
     0.404
    ],
    "easy-fact-02": [
     "c",
     0.343
    ],
    "easy-fact-03": [
     "c",
     0.419
    ],
    "easy-fact-04": [
     "c",
     0.38
    ],
    "easy-fact-05": [
     "c",
     0.361
    ],
    "easy-fact-06": [
     "c",
     0.368
    ],
    "easy-fact-07": [
     "c",
     0.485
    ],
    "easy-fact-08": [
     "c",
     0.375
    ],
    "easy-fact-09": [
     "c",
     0.376
    ],
    "easy-fact-10": [
     "c",
     0.408
    ],
    "easy-fact-11": [
     "c",
     0.41
    ],
    "easy-extraction-00": [
     "c",
     0.357
    ],
    "easy-extraction-01": [
     "c",
     0.339
    ],
    "easy-extraction-02": [
     "c",
     0.367
    ],
    "easy-extraction-03": [
     "c",
     0.384
    ],
    "easy-extraction-04": [
     "c",
     0.361
    ],
    "easy-extraction-05": [
     "c",
     0.318
    ],
    "easy-extraction-06": [
     "c",
     0.341
    ],
    "easy-extraction-07": [
     "c",
     0.369
    ],
    "easy-extraction-08": [
     "c",
     0.384
    ],
    "easy-extraction-09": [
     "c",
     0.342
    ],
    "easy-extraction-10": [
     "c",
     0.352
    ],
    "easy-extraction-11": [
     "c",
     0.401
    ],
    "easy-tool_selection-00": [
     "c",
     0.365
    ],
    "easy-tool_selection-01": [
     "c",
     0.407
    ],
    "easy-tool_selection-02": [
     "c",
     0.377
    ],
    "easy-tool_selection-03": [
     "c",
     0.398
    ],
    "easy-tool_selection-04": [
     "c",
     0.452
    ],
    "easy-tool_selection-05": [
     "c",
     0.336
    ],
    "easy-tool_selection-06": [
     "c",
     0.41
    ],
    "easy-tool_selection-07": [
     "c",
     0.334
    ],
    "easy-tool_selection-08": [
     "c",
     0.366
    ],
    "easy-tool_selection-09": [
     "c",
     0.364
    ],
    "easy-tool_selection-10": [
     "c",
     0.453
    ],
    "easy-tool_selection-11": [
     "c",
     0.371
    ],
    "original-policy-01-0": [
     "c",
     0.535
    ],
    "original-policy-01-1": [
     "c",
     0.402
    ],
    "original-policy-02-0": [
     "c",
     0.386
    ],
    "original-policy-02-1": [
     "c",
     0.405
    ],
    "original-policy-03-0": [
     "c",
     0.36
    ],
    "original-policy-03-1": [
     "c",
     0.347
    ],
    "original-policy-04-0": [
     "c",
     0.459
    ],
    "original-policy-04-1": [
     "c",
     0.755
    ],
    "original-policy-05-0": [
     "c",
     0.453
    ],
    "original-policy-05-1": [
     "c",
     0.344
    ],
    "original-policy-06-0": [
     "c",
     0.38
    ],
    "original-policy-06-1": [
     "w",
     0.374
    ],
    "original-intent-01-0": [
     "c",
     0.414
    ],
    "original-intent-01-1": [
     "c",
     0.348
    ],
    "original-intent-02-0": [
     "c",
     0.336
    ],
    "original-intent-02-1": [
     "c",
     0.367
    ],
    "original-intent-03-0": [
     "c",
     0.408
    ],
    "original-intent-03-1": [
     "c",
     0.382
    ],
    "original-intent-04-0": [
     "c",
     0.413
    ],
    "original-intent-04-1": [
     "c",
     0.358
    ],
    "original-intent-05-0": [
     "c",
     0.414
    ],
    "original-intent-05-1": [
     "c",
     0.402
    ],
    "original-intent-06-0": [
     "c",
     0.368
    ],
    "original-intent-06-1": [
     "c",
     0.383
    ],
    "original-ordinal-01-0": [
     "c",
     0.354
    ],
    "original-ordinal-01-1": [
     "c",
     0.409
    ],
    "original-ordinal-02-0": [
     "c",
     0.389
    ],
    "original-ordinal-02-1": [
     "c",
     0.415
    ],
    "original-ordinal-03-0": [
     "c",
     0.432
    ],
    "original-ordinal-03-1": [
     "c",
     0.387
    ],
    "original-ordinal-04-0": [
     "c",
     0.448
    ],
    "original-ordinal-04-1": [
     "c",
     0.4
    ],
    "original-ordinal-05-0": [
     "c",
     0.385
    ],
    "original-ordinal-05-1": [
     "c",
     0.385
    ],
    "original-ordinal-06-0": [
     "c",
     0.374
    ],
    "original-ordinal-06-1": [
     "c",
     0.399
    ],
    "original-extraction-01-0": [
     "c",
     0.404
    ],
    "original-extraction-01-1": [
     "c",
     0.394
    ],
    "original-extraction-02-0": [
     "c",
     0.404
    ],
    "original-extraction-02-1": [
     "c",
     0.369
    ],
    "original-extraction-03-0": [
     "c",
     0.365
    ],
    "original-extraction-03-1": [
     "c",
     0.364
    ],
    "original-extraction-04-0": [
     "c",
     0.428
    ],
    "original-extraction-04-1": [
     "c",
     0.429
    ],
    "original-extraction-05-0": [
     "c",
     0.349
    ],
    "original-extraction-05-1": [
     "c",
     0.403
    ],
    "original-extraction-06-0": [
     "c",
     0.348
    ],
    "original-extraction-06-1": [
     "c",
     0.402
    ],
    "original-adequacy-01-0": [
     "c",
     0.395
    ],
    "original-adequacy-01-1": [
     "c",
     0.419
    ],
    "original-adequacy-02-0": [
     "c",
     0.371
    ],
    "original-adequacy-02-1": [
     "c",
     0.392
    ],
    "original-adequacy-03-0": [
     "c",
     0.338
    ],
    "original-adequacy-03-1": [
     "c",
     0.368
    ],
    "original-adequacy-04-0": [
     "c",
     0.357
    ],
    "original-adequacy-04-1": [
     "c",
     0.41
    ],
    "original-adequacy-05-0": [
     "c",
     0.364
    ],
    "original-adequacy-05-1": [
     "c",
     0.398
    ],
    "original-adequacy-06-0": [
     "c",
     0.346
    ],
    "original-adequacy-06-1": [
     "c",
     0.358
    ],
    "original-routing-01-0": [
     "c",
     0.393
    ],
    "original-routing-01-1": [
     "c",
     0.369
    ],
    "original-routing-02-0": [
     "c",
     0.437
    ],
    "original-routing-02-1": [
     "c",
     0.351
    ],
    "original-routing-03-0": [
     "c",
     0.35
    ],
    "original-routing-03-1": [
     "c",
     0.372
    ],
    "original-routing-04-0": [
     "c",
     0.402
    ],
    "original-routing-04-1": [
     "c",
     0.426
    ],
    "original-routing-05-0": [
     "c",
     0.379
    ],
    "original-routing-05-1": [
     "c",
     0.338
    ],
    "original-routing-06-0": [
     "c",
     0.397
    ],
    "original-routing-06-1": [
     "c",
     0.347
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.399
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.394
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.373
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.386
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.427
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.424
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.361
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.456
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.366
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.319
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.349
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.465
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.369
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.354
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.386
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.358
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.34
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.327
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.34
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.374
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.36
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.449
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.492
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.417
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.39
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.398
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.395
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.33
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.439
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.366
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.437
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.366
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.411
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.417
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.356
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.373
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.328
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.368
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.325
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.443
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.372
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.362
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.457
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.399
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.35
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.462
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.463
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.39
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.404
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.351
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.409
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.377
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.585
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.356
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.389
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.377
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.346
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.377
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.405
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.442
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.366
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.372
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.367
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.374
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.406
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.382
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.356
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.381
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.438
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.35
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.354
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.357
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.382
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.374
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.35
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.346
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.396
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.338
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.342
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.426
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.482
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.443
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.361
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.364
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.363
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.401
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.436
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.334
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.382
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.354
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.317
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.326
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.412
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.448
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.322
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.361
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.394
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.333
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.373
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.392
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.388
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.321
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.371
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.379
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.41
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.381
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.409
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.357
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.372
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.369
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.404
    ]
   }
  },
  "decider-2b": {
   "display": "decider-2b (Mapika)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 82,
     "w": 14,
     "f": 0,
     "n": 0,
     "accuracy": 0.8542
    },
    "judge": {
     "c": 113,
     "w": 33,
     "f": 0,
     "n": 0,
     "accuracy": 0.774
    },
    "hard": {
     "c": 104,
     "w": 116,
     "f": 0,
     "n": 0,
     "accuracy": 0.4727
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "extraction": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "adequacy": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "routing": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     }
    },
    "judge": {
     "routing": {
      "c": 51,
      "w": 27,
      "f": 0,
      "n": 0,
      "accuracy": 0.6538
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 10,
      "w": 28,
      "f": 0,
      "n": 0,
      "accuracy": 0.2632
     },
     "probability": {
      "c": 6,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "temporal_numeric": {
      "c": 5,
      "w": 25,
      "f": 0,
      "n": 0,
      "accuracy": 0.1667
     },
     "ambiguous": {
      "c": 4,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.2857
     },
     "multi_hop": {
      "c": 17,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4857
     },
     "tradeoff": {
      "c": 6,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adversarial": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 21,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.6364
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.261
    ],
    "easy-intent-01": [
     "c",
     0.259
    ],
    "easy-intent-02": [
     "c",
     0.258
    ],
    "easy-intent-03": [
     "c",
     0.26
    ],
    "easy-intent-04": [
     "c",
     0.259
    ],
    "easy-intent-05": [
     "c",
     0.259
    ],
    "easy-intent-06": [
     "c",
     0.261
    ],
    "easy-intent-07": [
     "c",
     0.26
    ],
    "easy-intent-08": [
     "c",
     0.261
    ],
    "easy-intent-09": [
     "c",
     0.26
    ],
    "easy-intent-10": [
     "c",
     0.262
    ],
    "easy-intent-11": [
     "c",
     0.26
    ],
    "easy-fact-00": [
     "c",
     0.259
    ],
    "easy-fact-01": [
     "c",
     0.259
    ],
    "easy-fact-02": [
     "c",
     0.26
    ],
    "easy-fact-03": [
     "c",
     0.259
    ],
    "easy-fact-04": [
     "c",
     0.259
    ],
    "easy-fact-05": [
     "c",
     0.26
    ],
    "easy-fact-06": [
     "c",
     0.26
    ],
    "easy-fact-07": [
     "c",
     0.259
    ],
    "easy-fact-08": [
     "c",
     0.259
    ],
    "easy-fact-09": [
     "c",
     0.258
    ],
    "easy-fact-10": [
     "c",
     0.26
    ],
    "easy-fact-11": [
     "c",
     0.26
    ],
    "easy-extraction-00": [
     "c",
     0.259
    ],
    "easy-extraction-01": [
     "c",
     0.259
    ],
    "easy-extraction-02": [
     "c",
     0.26
    ],
    "easy-extraction-03": [
     "c",
     0.261
    ],
    "easy-extraction-04": [
     "c",
     0.259
    ],
    "easy-extraction-05": [
     "c",
     0.261
    ],
    "easy-extraction-06": [
     "c",
     0.26
    ],
    "easy-extraction-07": [
     "c",
     0.261
    ],
    "easy-extraction-08": [
     "c",
     0.258
    ],
    "easy-extraction-09": [
     "c",
     0.26
    ],
    "easy-extraction-10": [
     "c",
     0.258
    ],
    "easy-extraction-11": [
     "c",
     0.264
    ],
    "easy-tool_selection-00": [
     "c",
     0.26
    ],
    "easy-tool_selection-01": [
     "c",
     0.259
    ],
    "easy-tool_selection-02": [
     "c",
     0.26
    ],
    "easy-tool_selection-03": [
     "c",
     0.26
    ],
    "easy-tool_selection-04": [
     "c",
     0.26
    ],
    "easy-tool_selection-05": [
     "c",
     0.259
    ],
    "easy-tool_selection-06": [
     "c",
     0.259
    ],
    "easy-tool_selection-07": [
     "c",
     0.261
    ],
    "easy-tool_selection-08": [
     "c",
     0.259
    ],
    "easy-tool_selection-09": [
     "c",
     0.26
    ],
    "easy-tool_selection-10": [
     "c",
     0.26
    ],
    "easy-tool_selection-11": [
     "c",
     0.259
    ],
    "original-policy-01-0": [
     "c",
     0.261
    ],
    "original-policy-01-1": [
     "c",
     0.26
    ],
    "original-policy-02-0": [
     "c",
     0.259
    ],
    "original-policy-02-1": [
     "w",
     0.26
    ],
    "original-policy-03-0": [
     "w",
     0.258
    ],
    "original-policy-03-1": [
     "c",
     0.258
    ],
    "original-policy-04-0": [
     "c",
     0.259
    ],
    "original-policy-04-1": [
     "c",
     0.259
    ],
    "original-policy-05-0": [
     "c",
     0.259
    ],
    "original-policy-05-1": [
     "c",
     0.258
    ],
    "original-policy-06-0": [
     "c",
     0.26
    ],
    "original-policy-06-1": [
     "c",
     0.261
    ],
    "original-intent-01-0": [
     "c",
     0.26
    ],
    "original-intent-01-1": [
     "c",
     0.258
    ],
    "original-intent-02-0": [
     "c",
     0.259
    ],
    "original-intent-02-1": [
     "c",
     0.259
    ],
    "original-intent-03-0": [
     "c",
     0.259
    ],
    "original-intent-03-1": [
     "c",
     0.258
    ],
    "original-intent-04-0": [
     "c",
     0.259
    ],
    "original-intent-04-1": [
     "c",
     0.259
    ],
    "original-intent-05-0": [
     "c",
     0.26
    ],
    "original-intent-05-1": [
     "c",
     0.26
    ],
    "original-intent-06-0": [
     "c",
     0.259
    ],
    "original-intent-06-1": [
     "c",
     0.259
    ],
    "original-ordinal-01-0": [
     "c",
     0.259
    ],
    "original-ordinal-01-1": [
     "c",
     0.259
    ],
    "original-ordinal-02-0": [
     "c",
     0.26
    ],
    "original-ordinal-02-1": [
     "c",
     0.259
    ],
    "original-ordinal-03-0": [
     "c",
     0.26
    ],
    "original-ordinal-03-1": [
     "c",
     0.259
    ],
    "original-ordinal-04-0": [
     "c",
     0.259
    ],
    "original-ordinal-04-1": [
     "c",
     0.259
    ],
    "original-ordinal-05-0": [
     "c",
     0.26
    ],
    "original-ordinal-05-1": [
     "w",
     0.26
    ],
    "original-ordinal-06-0": [
     "c",
     0.26
    ],
    "original-ordinal-06-1": [
     "c",
     0.26
    ],
    "original-extraction-01-0": [
     "c",
     0.259
    ],
    "original-extraction-01-1": [
     "c",
     0.258
    ],
    "original-extraction-02-0": [
     "c",
     0.268
    ],
    "original-extraction-02-1": [
     "c",
     0.26
    ],
    "original-extraction-03-0": [
     "c",
     0.258
    ],
    "original-extraction-03-1": [
     "c",
     0.259
    ],
    "original-extraction-04-0": [
     "c",
     0.26
    ],
    "original-extraction-04-1": [
     "c",
     0.259
    ],
    "original-extraction-05-0": [
     "c",
     0.259
    ],
    "original-extraction-05-1": [
     "c",
     0.26
    ],
    "original-extraction-06-0": [
     "c",
     0.26
    ],
    "original-extraction-06-1": [
     "c",
     0.259
    ],
    "original-adequacy-01-0": [
     "c",
     0.258
    ],
    "original-adequacy-01-1": [
     "c",
     0.258
    ],
    "original-adequacy-02-0": [
     "c",
     0.258
    ],
    "original-adequacy-02-1": [
     "c",
     0.258
    ],
    "original-adequacy-03-0": [
     "w",
     0.259
    ],
    "original-adequacy-03-1": [
     "w",
     0.258
    ],
    "original-adequacy-04-0": [
     "c",
     0.258
    ],
    "original-adequacy-04-1": [
     "c",
     0.258
    ],
    "original-adequacy-05-0": [
     "w",
     0.26
    ],
    "original-adequacy-05-1": [
     "w",
     0.26
    ],
    "original-adequacy-06-0": [
     "c",
     0.258
    ],
    "original-adequacy-06-1": [
     "w",
     0.26
    ],
    "original-routing-01-0": [
     "c",
     0.26
    ],
    "original-routing-01-1": [
     "c",
     0.261
    ],
    "original-routing-02-0": [
     "c",
     0.261
    ],
    "original-routing-02-1": [
     "c",
     0.258
    ],
    "original-routing-03-0": [
     "c",
     0.26
    ],
    "original-routing-03-1": [
     "c",
     0.261
    ],
    "original-routing-04-0": [
     "w",
     0.259
    ],
    "original-routing-04-1": [
     "c",
     0.259
    ],
    "original-routing-05-0": [
     "w",
     0.26
    ],
    "original-routing-05-1": [
     "w",
     0.259
    ],
    "original-routing-06-0": [
     "c",
     0.259
    ],
    "original-routing-06-1": [
     "c",
     0.259
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.474
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.013
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     0.314
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.312
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.313
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.312
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.317
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.311
    ],
    "hard-opus-a-probability-03": [
     "w",
     0.264
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.278
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.279
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.264
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.265
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.265
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     0.283
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.266
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.267
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.268
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.266
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.266
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.269
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.267
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.265
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.266
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.336
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.319
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     0.311
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.311
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     0.311
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.289
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.265
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.277
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.268
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.273
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.274
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.27
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.274
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.27
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.268
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.269
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.468
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.454
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.463
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.455
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.456
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     0.454
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.454
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.265
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.271
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.272
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.271
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.274
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.274
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.275
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.263
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.261
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.263
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.262
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.261
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.264
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.332
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.313
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.312
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.312
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     0.312
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.319
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.322
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.312
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.262
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.26
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.264
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.263
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.262
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.262
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.261
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.264
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.262
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.261
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.261
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.261
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.261
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.262
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.262
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.26
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.457
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.44
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     0.433
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.434
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.261
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.263
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.262
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.264
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.263
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.262
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.264
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.262
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.263
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.265
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.264
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.264
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.264
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.264
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.264
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.265
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.263
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.263
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.333
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.318
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.319
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.266
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.329
    ]
   }
  },
  "decider-35b-a3b": {
   "display": "decider-35b-a3b (Mapika)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 93,
     "w": 3,
     "f": 0,
     "n": 0,
     "accuracy": 0.9688
    },
    "judge": {
     "c": 133,
     "w": 13,
     "f": 0,
     "n": 0,
     "accuracy": 0.911
    },
    "hard": {
     "c": 144,
     "w": 76,
     "f": 0,
     "n": 0,
     "accuracy": 0.6545
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 70,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8974
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 20,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.5263
     },
     "probability": {
      "c": 12,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.6
     },
     "temporal_numeric": {
      "c": 6,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     },
     "ambiguous": {
      "c": 9,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6429
     },
     "multi_hop": {
      "c": 26,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.7429
     },
     "tradeoff": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 25,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.7576
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     2.969
    ],
    "easy-intent-01": [
     "c",
     0.284
    ],
    "easy-intent-02": [
     "c",
     0.284
    ],
    "easy-intent-03": [
     "c",
     0.284
    ],
    "easy-intent-04": [
     "c",
     0.283
    ],
    "easy-intent-05": [
     "c",
     0.284
    ],
    "easy-intent-06": [
     "c",
     0.283
    ],
    "easy-intent-07": [
     "c",
     0.283
    ],
    "easy-intent-08": [
     "c",
     0.283
    ],
    "easy-intent-09": [
     "c",
     0.284
    ],
    "easy-intent-10": [
     "c",
     0.284
    ],
    "easy-intent-11": [
     "c",
     0.283
    ],
    "easy-fact-00": [
     "c",
     0.284
    ],
    "easy-fact-01": [
     "c",
     0.284
    ],
    "easy-fact-02": [
     "c",
     0.282
    ],
    "easy-fact-03": [
     "c",
     0.284
    ],
    "easy-fact-04": [
     "c",
     0.283
    ],
    "easy-fact-05": [
     "c",
     0.283
    ],
    "easy-fact-06": [
     "c",
     0.284
    ],
    "easy-fact-07": [
     "c",
     0.283
    ],
    "easy-fact-08": [
     "c",
     2.133
    ],
    "easy-fact-09": [
     "c",
     0.279
    ],
    "easy-fact-10": [
     "c",
     0.281
    ],
    "easy-fact-11": [
     "c",
     0.282
    ],
    "easy-extraction-00": [
     "c",
     0.28
    ],
    "easy-extraction-01": [
     "c",
     0.284
    ],
    "easy-extraction-02": [
     "c",
     0.282
    ],
    "easy-extraction-03": [
     "c",
     0.283
    ],
    "easy-extraction-04": [
     "c",
     0.282
    ],
    "easy-extraction-05": [
     "c",
     0.283
    ],
    "easy-extraction-06": [
     "c",
     0.281
    ],
    "easy-extraction-07": [
     "c",
     0.281
    ],
    "easy-extraction-08": [
     "c",
     0.283
    ],
    "easy-extraction-09": [
     "c",
     0.282
    ],
    "easy-extraction-10": [
     "c",
     0.28
    ],
    "easy-extraction-11": [
     "c",
     0.282
    ],
    "easy-tool_selection-00": [
     "c",
     0.284
    ],
    "easy-tool_selection-01": [
     "c",
     0.285
    ],
    "easy-tool_selection-02": [
     "c",
     0.285
    ],
    "easy-tool_selection-03": [
     "c",
     0.285
    ],
    "easy-tool_selection-04": [
     "c",
     0.283
    ],
    "easy-tool_selection-05": [
     "c",
     0.284
    ],
    "easy-tool_selection-06": [
     "c",
     0.285
    ],
    "easy-tool_selection-07": [
     "c",
     0.284
    ],
    "easy-tool_selection-08": [
     "c",
     0.283
    ],
    "easy-tool_selection-09": [
     "c",
     0.284
    ],
    "easy-tool_selection-10": [
     "c",
     0.284
    ],
    "easy-tool_selection-11": [
     "c",
     0.285
    ],
    "original-policy-01-0": [
     "c",
     0.286
    ],
    "original-policy-01-1": [
     "c",
     0.285
    ],
    "original-policy-02-0": [
     "c",
     0.283
    ],
    "original-policy-02-1": [
     "c",
     0.283
    ],
    "original-policy-03-0": [
     "c",
     0.283
    ],
    "original-policy-03-1": [
     "c",
     0.283
    ],
    "original-policy-04-0": [
     "c",
     0.281
    ],
    "original-policy-04-1": [
     "c",
     0.282
    ],
    "original-policy-05-0": [
     "w",
     0.284
    ],
    "original-policy-05-1": [
     "c",
     0.283
    ],
    "original-policy-06-0": [
     "c",
     0.283
    ],
    "original-policy-06-1": [
     "c",
     0.284
    ],
    "original-intent-01-0": [
     "c",
     0.281
    ],
    "original-intent-01-1": [
     "c",
     0.282
    ],
    "original-intent-02-0": [
     "c",
     0.281
    ],
    "original-intent-02-1": [
     "c",
     0.28
    ],
    "original-intent-03-0": [
     "c",
     0.282
    ],
    "original-intent-03-1": [
     "c",
     0.282
    ],
    "original-intent-04-0": [
     "c",
     0.283
    ],
    "original-intent-04-1": [
     "c",
     0.282
    ],
    "original-intent-05-0": [
     "c",
     0.283
    ],
    "original-intent-05-1": [
     "c",
     0.282
    ],
    "original-intent-06-0": [
     "c",
     0.282
    ],
    "original-intent-06-1": [
     "c",
     0.282
    ],
    "original-ordinal-01-0": [
     "c",
     0.282
    ],
    "original-ordinal-01-1": [
     "c",
     0.283
    ],
    "original-ordinal-02-0": [
     "c",
     0.283
    ],
    "original-ordinal-02-1": [
     "c",
     0.282
    ],
    "original-ordinal-03-0": [
     "c",
     0.282
    ],
    "original-ordinal-03-1": [
     "c",
     0.283
    ],
    "original-ordinal-04-0": [
     "c",
     0.283
    ],
    "original-ordinal-04-1": [
     "c",
     0.284
    ],
    "original-ordinal-05-0": [
     "c",
     0.285
    ],
    "original-ordinal-05-1": [
     "c",
     0.285
    ],
    "original-ordinal-06-0": [
     "c",
     0.286
    ],
    "original-ordinal-06-1": [
     "c",
     0.285
    ],
    "original-extraction-01-0": [
     "c",
     0.282
    ],
    "original-extraction-01-1": [
     "c",
     0.285
    ],
    "original-extraction-02-0": [
     "c",
     0.282
    ],
    "original-extraction-02-1": [
     "c",
     0.282
    ],
    "original-extraction-03-0": [
     "c",
     0.283
    ],
    "original-extraction-03-1": [
     "c",
     0.282
    ],
    "original-extraction-04-0": [
     "c",
     0.282
    ],
    "original-extraction-04-1": [
     "c",
     0.283
    ],
    "original-extraction-05-0": [
     "c",
     0.281
    ],
    "original-extraction-05-1": [
     "c",
     0.282
    ],
    "original-extraction-06-0": [
     "c",
     0.283
    ],
    "original-extraction-06-1": [
     "c",
     0.283
    ],
    "original-adequacy-01-0": [
     "c",
     0.282
    ],
    "original-adequacy-01-1": [
     "c",
     0.282
    ],
    "original-adequacy-02-0": [
     "c",
     0.282
    ],
    "original-adequacy-02-1": [
     "c",
     0.283
    ],
    "original-adequacy-03-0": [
     "c",
     0.283
    ],
    "original-adequacy-03-1": [
     "c",
     0.283
    ],
    "original-adequacy-04-0": [
     "c",
     0.283
    ],
    "original-adequacy-04-1": [
     "c",
     0.283
    ],
    "original-adequacy-05-0": [
     "w",
     0.281
    ],
    "original-adequacy-05-1": [
     "c",
     0.281
    ],
    "original-adequacy-06-0": [
     "c",
     0.283
    ],
    "original-adequacy-06-1": [
     "c",
     0.283
    ],
    "original-routing-01-0": [
     "c",
     0.286
    ],
    "original-routing-01-1": [
     "c",
     0.285
    ],
    "original-routing-02-0": [
     "c",
     0.285
    ],
    "original-routing-02-1": [
     "c",
     0.283
    ],
    "original-routing-03-0": [
     "c",
     0.284
    ],
    "original-routing-03-1": [
     "c",
     0.285
    ],
    "original-routing-04-0": [
     "c",
     0.284
    ],
    "original-routing-04-1": [
     "c",
     0.286
    ],
    "original-routing-05-0": [
     "c",
     0.284
    ],
    "original-routing-05-1": [
     "c",
     0.285
    ],
    "original-routing-06-0": [
     "c",
     0.284
    ],
    "original-routing-06-1": [
     "c",
     0.284
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.623
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.126
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.445
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.445
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.445
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     0.444
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.444
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.444
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.303
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.343
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.344
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.298
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.304
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.304
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.358
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.302
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.312
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.312
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.304
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.305
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.312
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.303
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     0.303
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.313
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.444
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.444
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     0.446
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.446
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.445
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.385
    ],
    "hard-opus-b-probability-02": [
     "c",
     0.305
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.345
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.312
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.33
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.328
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.318
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.329
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.318
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.312
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.311
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.621
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     0.623
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.623
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.621
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.62
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.623
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.622
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.311
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.328
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.331
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.318
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.331
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.33
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.329
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.292
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.289
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.295
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.291
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.291
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.296
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.444
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.443
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     0.444
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.447
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.444
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.446
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.446
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     0.446
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.295
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.288
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.297
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.298
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.295
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.288
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.29
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.296
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.297
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.292
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.292
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.293
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.29
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.292
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.291
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.288
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     0.568
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     0.57
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.566
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     0.568
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.296
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.295
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.296
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.295
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.296
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.295
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.295
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.291
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.292
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.299
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.299
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.299
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.296
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.299
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.3
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.299
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.297
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.295
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.444
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.446
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.447
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.311
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.445
    ]
   }
  },
  "decision-machine-1": {
   "display": "decision-machine-1 (milliseconds.ai)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 73,
     "w": 23,
     "f": 0,
     "n": 0,
     "accuracy": 0.7604
    },
    "judge": {
     "c": 131,
     "w": 15,
     "f": 0,
     "n": 0,
     "accuracy": 0.8973
    },
    "hard": {
     "c": 103,
     "w": 117,
     "f": 0,
     "n": 0,
     "accuracy": 0.4682
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "intent": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "adequacy": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "routing": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     }
    },
    "judge": {
     "routing": {
      "c": 69,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.8846
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 13,
      "w": 25,
      "f": 0,
      "n": 0,
      "accuracy": 0.3421
     },
     "probability": {
      "c": 9,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.45
     },
     "temporal_numeric": {
      "c": 7,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.2333
     },
     "ambiguous": {
      "c": 3,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.2143
     },
     "multi_hop": {
      "c": 16,
      "w": 19,
      "f": 0,
      "n": 0,
      "accuracy": 0.4571
     },
     "tradeoff": {
      "c": 4,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "judge_hard": {
      "c": 21,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.6364
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.449
    ],
    "easy-intent-01": [
     "c",
     0.149
    ],
    "easy-intent-02": [
     "c",
     0.198
    ],
    "easy-intent-03": [
     "c",
     0.145
    ],
    "easy-intent-04": [
     "c",
     0.15
    ],
    "easy-intent-05": [
     "c",
     0.198
    ],
    "easy-intent-06": [
     "c",
     0.161
    ],
    "easy-intent-07": [
     "c",
     0.164
    ],
    "easy-intent-08": [
     "c",
     0.196
    ],
    "easy-intent-09": [
     "c",
     0.152
    ],
    "easy-intent-10": [
     "c",
     0.158
    ],
    "easy-intent-11": [
     "c",
     0.192
    ],
    "easy-fact-00": [
     "c",
     0.146
    ],
    "easy-fact-01": [
     "c",
     0.358
    ],
    "easy-fact-02": [
     "c",
     0.184
    ],
    "easy-fact-03": [
     "c",
     0.183
    ],
    "easy-fact-04": [
     "c",
     0.154
    ],
    "easy-fact-05": [
     "c",
     0.189
    ],
    "easy-fact-06": [
     "c",
     0.17
    ],
    "easy-fact-07": [
     "c",
     0.157
    ],
    "easy-fact-08": [
     "c",
     0.198
    ],
    "easy-fact-09": [
     "c",
     0.159
    ],
    "easy-fact-10": [
     "c",
     0.152
    ],
    "easy-fact-11": [
     "c",
     0.187
    ],
    "easy-extraction-00": [
     "c",
     0.162
    ],
    "easy-extraction-01": [
     "c",
     0.17
    ],
    "easy-extraction-02": [
     "c",
     0.208
    ],
    "easy-extraction-03": [
     "c",
     0.149
    ],
    "easy-extraction-04": [
     "c",
     0.15
    ],
    "easy-extraction-05": [
     "c",
     0.193
    ],
    "easy-extraction-06": [
     "c",
     0.165
    ],
    "easy-extraction-07": [
     "c",
     0.157
    ],
    "easy-extraction-08": [
     "c",
     0.183
    ],
    "easy-extraction-09": [
     "c",
     0.145
    ],
    "easy-extraction-10": [
     "c",
     0.146
    ],
    "easy-extraction-11": [
     "c",
     0.197
    ],
    "easy-tool_selection-00": [
     "c",
     0.153
    ],
    "easy-tool_selection-01": [
     "c",
     0.159
    ],
    "easy-tool_selection-02": [
     "c",
     0.2
    ],
    "easy-tool_selection-03": [
     "c",
     0.155
    ],
    "easy-tool_selection-04": [
     "c",
     0.16
    ],
    "easy-tool_selection-05": [
     "c",
     0.187
    ],
    "easy-tool_selection-06": [
     "c",
     0.146
    ],
    "easy-tool_selection-07": [
     "c",
     0.159
    ],
    "easy-tool_selection-08": [
     "c",
     0.188
    ],
    "easy-tool_selection-09": [
     "c",
     0.151
    ],
    "easy-tool_selection-10": [
     "c",
     0.153
    ],
    "easy-tool_selection-11": [
     "c",
     0.208
    ],
    "original-policy-01-0": [
     "c",
     0.166
    ],
    "original-policy-01-1": [
     "c",
     0.154
    ],
    "original-policy-02-0": [
     "w",
     0.196
    ],
    "original-policy-02-1": [
     "w",
     0.15
    ],
    "original-policy-03-0": [
     "w",
     0.151
    ],
    "original-policy-03-1": [
     "w",
     0.222
    ],
    "original-policy-04-0": [
     "w",
     0.18
    ],
    "original-policy-04-1": [
     "c",
     0.166
    ],
    "original-policy-05-0": [
     "c",
     0.191
    ],
    "original-policy-05-1": [
     "c",
     0.166
    ],
    "original-policy-06-0": [
     "c",
     0.159
    ],
    "original-policy-06-1": [
     "c",
     0.186
    ],
    "original-intent-01-0": [
     "c",
     0.161
    ],
    "original-intent-01-1": [
     "c",
     0.154
    ],
    "original-intent-02-0": [
     "w",
     0.182
    ],
    "original-intent-02-1": [
     "c",
     0.145
    ],
    "original-intent-03-0": [
     "c",
     0.147
    ],
    "original-intent-03-1": [
     "c",
     0.203
    ],
    "original-intent-04-0": [
     "c",
     0.18
    ],
    "original-intent-04-1": [
     "c",
     0.156
    ],
    "original-intent-05-0": [
     "w",
     0.183
    ],
    "original-intent-05-1": [
     "w",
     0.149
    ],
    "original-intent-06-0": [
     "c",
     0.159
    ],
    "original-intent-06-1": [
     "c",
     0.411
    ],
    "original-ordinal-01-0": [
     "c",
     0.154
    ],
    "original-ordinal-01-1": [
     "c",
     0.15
    ],
    "original-ordinal-02-0": [
     "c",
     0.19
    ],
    "original-ordinal-02-1": [
     "c",
     0.157
    ],
    "original-ordinal-03-0": [
     "c",
     0.158
    ],
    "original-ordinal-03-1": [
     "c",
     0.198
    ],
    "original-ordinal-04-0": [
     "c",
     0.151
    ],
    "original-ordinal-04-1": [
     "c",
     0.145
    ],
    "original-ordinal-05-0": [
     "c",
     0.192
    ],
    "original-ordinal-05-1": [
     "c",
     0.168
    ],
    "original-ordinal-06-0": [
     "c",
     0.148
    ],
    "original-ordinal-06-1": [
     "c",
     0.182
    ],
    "original-extraction-01-0": [
     "c",
     0.147
    ],
    "original-extraction-01-1": [
     "c",
     0.154
    ],
    "original-extraction-02-0": [
     "c",
     0.187
    ],
    "original-extraction-02-1": [
     "c",
     0.148
    ],
    "original-extraction-03-0": [
     "c",
     0.486
    ],
    "original-extraction-03-1": [
     "c",
     0.17
    ],
    "original-extraction-04-0": [
     "c",
     0.151
    ],
    "original-extraction-04-1": [
     "c",
     0.181
    ],
    "original-extraction-05-0": [
     "c",
     0.145
    ],
    "original-extraction-05-1": [
     "c",
     0.181
    ],
    "original-extraction-06-0": [
     "c",
     0.195
    ],
    "original-extraction-06-1": [
     "c",
     0.146
    ],
    "original-adequacy-01-0": [
     "c",
     0.359
    ],
    "original-adequacy-01-1": [
     "c",
     0.219
    ],
    "original-adequacy-02-0": [
     "c",
     0.153
    ],
    "original-adequacy-02-1": [
     "c",
     0.149
    ],
    "original-adequacy-03-0": [
     "w",
     0.199
    ],
    "original-adequacy-03-1": [
     "w",
     0.154
    ],
    "original-adequacy-04-0": [
     "c",
     0.169
    ],
    "original-adequacy-04-1": [
     "w",
     0.186
    ],
    "original-adequacy-05-0": [
     "c",
     0.146
    ],
    "original-adequacy-05-1": [
     "w",
     0.151
    ],
    "original-adequacy-06-0": [
     "c",
     0.196
    ],
    "original-adequacy-06-1": [
     "c",
     0.174
    ],
    "original-routing-01-0": [
     "c",
     0.148
    ],
    "original-routing-01-1": [
     "c",
     0.183
    ],
    "original-routing-02-0": [
     "c",
     0.161
    ],
    "original-routing-02-1": [
     "c",
     0.147
    ],
    "original-routing-03-0": [
     "c",
     0.194
    ],
    "original-routing-03-1": [
     "c",
     0.177
    ],
    "original-routing-04-0": [
     "w",
     0.153
    ],
    "original-routing-04-1": [
     "w",
     0.193
    ],
    "original-routing-05-0": [
     "w",
     0.169
    ],
    "original-routing-05-1": [
     "w",
     0.156
    ],
    "original-routing-06-0": [
     "w",
     0.187
    ],
    "original-routing-06-1": [
     "w",
     0.16
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.409
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.289
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.279
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.332
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.279
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.28
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.274
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.284
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.199
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.179
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.189
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.183
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.17
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.163
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.183
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.155
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.154
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.186
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.161
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.155
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.272
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.152
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.149
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.21
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.341
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.282
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.3
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.285
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     0.276
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.226
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.15
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.212
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.184
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.165
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.147
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.196
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.178
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.156
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.185
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.155
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.388
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.35
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.357
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.794
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.437
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.36
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.405
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.183
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.184
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.173
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.188
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.18
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.178
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.183
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.151
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.175
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.223
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.157
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.158
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.202
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.287
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.304
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     0.279
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.274
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.36
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.314
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.287
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.387
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.163
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.189
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.169
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.149
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.184
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.161
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.154
    ],
    "hard-sol-a-trap-15": [
     "w",
     0.182
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.169
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.171
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.152
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.162
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.2
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.144
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.147
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.195
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.486
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.303
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.293
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.294
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.176
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.215
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.289
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.204
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.163
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.165
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.181
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.158
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.163
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.195
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.156
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.145
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.184
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.159
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.157
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.193
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.164
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.151
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.299
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.272
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.476
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.152
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.276
    ]
   }
  },
  "djev": {
   "display": "djev (Maisa, diffusion-gemma)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 94,
     "w": 2,
     "f": 0,
     "n": 0,
     "accuracy": 0.9792
    },
    "judge": {
     "c": 136,
     "w": 10,
     "f": 0,
     "n": 0,
     "accuracy": 0.9315
    },
    "hard": {
     "c": 153,
     "w": 67,
     "f": 0,
     "n": 0,
     "accuracy": 0.6955
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 73,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9359
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 18,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.4737
     },
     "probability": {
      "c": 12,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.6
     },
     "temporal_numeric": {
      "c": 10,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "ambiguous": {
      "c": 10,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.7143
     },
     "multi_hop": {
      "c": 29,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.8286
     },
     "tradeoff": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 29,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.8788
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.109
    ],
    "easy-intent-01": [
     "c",
     0.257
    ],
    "easy-intent-02": [
     "c",
     0.216
    ],
    "easy-intent-03": [
     "c",
     0.243
    ],
    "easy-intent-04": [
     "c",
     0.225
    ],
    "easy-intent-05": [
     "c",
     0.232
    ],
    "easy-intent-06": [
     "c",
     0.233
    ],
    "easy-intent-07": [
     "c",
     0.224
    ],
    "easy-intent-08": [
     "c",
     0.212
    ],
    "easy-intent-09": [
     "c",
     0.224
    ],
    "easy-intent-10": [
     "c",
     0.236
    ],
    "easy-intent-11": [
     "c",
     0.223
    ],
    "easy-fact-00": [
     "c",
     0.243
    ],
    "easy-fact-01": [
     "c",
     0.244
    ],
    "easy-fact-02": [
     "c",
     0.246
    ],
    "easy-fact-03": [
     "c",
     0.232
    ],
    "easy-fact-04": [
     "c",
     0.234
    ],
    "easy-fact-05": [
     "c",
     0.217
    ],
    "easy-fact-06": [
     "c",
     0.238
    ],
    "easy-fact-07": [
     "c",
     0.253
    ],
    "easy-fact-08": [
     "c",
     0.228
    ],
    "easy-fact-09": [
     "c",
     0.243
    ],
    "easy-fact-10": [
     "c",
     0.228
    ],
    "easy-fact-11": [
     "c",
     0.232
    ],
    "easy-extraction-00": [
     "c",
     0.239
    ],
    "easy-extraction-01": [
     "c",
     0.239
    ],
    "easy-extraction-02": [
     "c",
     0.226
    ],
    "easy-extraction-03": [
     "c",
     0.242
    ],
    "easy-extraction-04": [
     "c",
     0.232
    ],
    "easy-extraction-05": [
     "c",
     0.213
    ],
    "easy-extraction-06": [
     "c",
     0.231
    ],
    "easy-extraction-07": [
     "c",
     0.24
    ],
    "easy-extraction-08": [
     "c",
     0.228
    ],
    "easy-extraction-09": [
     "c",
     0.224
    ],
    "easy-extraction-10": [
     "c",
     0.218
    ],
    "easy-extraction-11": [
     "c",
     0.237
    ],
    "easy-tool_selection-00": [
     "c",
     0.22
    ],
    "easy-tool_selection-01": [
     "c",
     0.23
    ],
    "easy-tool_selection-02": [
     "c",
     0.24
    ],
    "easy-tool_selection-03": [
     "c",
     0.28
    ],
    "easy-tool_selection-04": [
     "c",
     0.215
    ],
    "easy-tool_selection-05": [
     "c",
     0.218
    ],
    "easy-tool_selection-06": [
     "c",
     0.216
    ],
    "easy-tool_selection-07": [
     "c",
     0.224
    ],
    "easy-tool_selection-08": [
     "c",
     0.224
    ],
    "easy-tool_selection-09": [
     "c",
     0.244
    ],
    "easy-tool_selection-10": [
     "c",
     0.234
    ],
    "easy-tool_selection-11": [
     "c",
     0.247
    ],
    "original-policy-01-0": [
     "c",
     1.43
    ],
    "original-policy-01-1": [
     "c",
     0.345
    ],
    "original-policy-02-0": [
     "c",
     0.22
    ],
    "original-policy-02-1": [
     "c",
     0.309
    ],
    "original-policy-03-0": [
     "c",
     0.243
    ],
    "original-policy-03-1": [
     "c",
     0.224
    ],
    "original-policy-04-0": [
     "c",
     0.218
    ],
    "original-policy-04-1": [
     "c",
     0.29
    ],
    "original-policy-05-0": [
     "c",
     0.233
    ],
    "original-policy-05-1": [
     "c",
     0.227
    ],
    "original-policy-06-0": [
     "c",
     0.234
    ],
    "original-policy-06-1": [
     "c",
     0.226
    ],
    "original-intent-01-0": [
     "c",
     0.226
    ],
    "original-intent-01-1": [
     "c",
     0.228
    ],
    "original-intent-02-0": [
     "c",
     0.236
    ],
    "original-intent-02-1": [
     "c",
     0.228
    ],
    "original-intent-03-0": [
     "c",
     0.234
    ],
    "original-intent-03-1": [
     "c",
     0.239
    ],
    "original-intent-04-0": [
     "c",
     0.222
    ],
    "original-intent-04-1": [
     "c",
     0.24
    ],
    "original-intent-05-0": [
     "c",
     0.239
    ],
    "original-intent-05-1": [
     "c",
     0.225
    ],
    "original-intent-06-0": [
     "c",
     0.245
    ],
    "original-intent-06-1": [
     "c",
     0.234
    ],
    "original-ordinal-01-0": [
     "c",
     0.235
    ],
    "original-ordinal-01-1": [
     "c",
     0.22
    ],
    "original-ordinal-02-0": [
     "c",
     0.236
    ],
    "original-ordinal-02-1": [
     "c",
     0.231
    ],
    "original-ordinal-03-0": [
     "c",
     0.235
    ],
    "original-ordinal-03-1": [
     "c",
     0.247
    ],
    "original-ordinal-04-0": [
     "c",
     0.229
    ],
    "original-ordinal-04-1": [
     "c",
     0.213
    ],
    "original-ordinal-05-0": [
     "c",
     0.222
    ],
    "original-ordinal-05-1": [
     "c",
     0.307
    ],
    "original-ordinal-06-0": [
     "c",
     0.237
    ],
    "original-ordinal-06-1": [
     "c",
     0.22
    ],
    "original-extraction-01-0": [
     "c",
     0.295
    ],
    "original-extraction-01-1": [
     "c",
     0.238
    ],
    "original-extraction-02-0": [
     "c",
     0.232
    ],
    "original-extraction-02-1": [
     "c",
     0.219
    ],
    "original-extraction-03-0": [
     "c",
     0.379
    ],
    "original-extraction-03-1": [
     "c",
     0.248
    ],
    "original-extraction-04-0": [
     "c",
     0.235
    ],
    "original-extraction-04-1": [
     "c",
     0.215
    ],
    "original-extraction-05-0": [
     "c",
     0.3
    ],
    "original-extraction-05-1": [
     "c",
     0.226
    ],
    "original-extraction-06-0": [
     "c",
     0.237
    ],
    "original-extraction-06-1": [
     "c",
     0.242
    ],
    "original-adequacy-01-0": [
     "c",
     0.24
    ],
    "original-adequacy-01-1": [
     "c",
     0.223
    ],
    "original-adequacy-02-0": [
     "c",
     0.241
    ],
    "original-adequacy-02-1": [
     "c",
     0.243
    ],
    "original-adequacy-03-0": [
     "c",
     0.226
    ],
    "original-adequacy-03-1": [
     "w",
     0.224
    ],
    "original-adequacy-04-0": [
     "c",
     0.237
    ],
    "original-adequacy-04-1": [
     "c",
     0.227
    ],
    "original-adequacy-05-0": [
     "c",
     0.244
    ],
    "original-adequacy-05-1": [
     "c",
     0.234
    ],
    "original-adequacy-06-0": [
     "c",
     0.25
    ],
    "original-adequacy-06-1": [
     "c",
     0.242
    ],
    "original-routing-01-0": [
     "c",
     0.227
    ],
    "original-routing-01-1": [
     "c",
     0.237
    ],
    "original-routing-02-0": [
     "c",
     0.226
    ],
    "original-routing-02-1": [
     "c",
     0.228
    ],
    "original-routing-03-0": [
     "c",
     0.256
    ],
    "original-routing-03-1": [
     "c",
     0.23
    ],
    "original-routing-04-0": [
     "c",
     0.235
    ],
    "original-routing-04-1": [
     "c",
     0.222
    ],
    "original-routing-05-0": [
     "c",
     0.227
    ],
    "original-routing-05-1": [
     "c",
     0.225
    ],
    "original-routing-06-0": [
     "c",
     0.227
    ],
    "original-routing-06-1": [
     "c",
     0.229
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     1.452
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.383
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.293
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.316
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.296
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.292
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.301
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.309
    ],
    "hard-opus-a-probability-03": [
     "w",
     0.231
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.259
    ],
    "hard-opus-a-probability-07": [
     "c",
     0.254
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.244
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.254
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.233
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.263
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.246
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     0.249
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.247
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.248
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.224
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.245
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.235
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.236
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.247
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     0.321
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     0.299
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.398
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.331
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.329
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.28
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.232
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.26
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.239
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.252
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.242
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.255
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.25
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.255
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.242
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.261
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.37
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.348
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.334
    ],
    "hard-opus-c-long_policy-05": [
     "c",
     0.354
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.318
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.404
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     0.356
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.229
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.238
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.239
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.252
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.247
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.241
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.24
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.241
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.23
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.244
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.238
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.255
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.235
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.306
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.307
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.315
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     0.305
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.289
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.297
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.306
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.304
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.238
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.235
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.237
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.252
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.231
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.237
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.221
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.244
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.233
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.227
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.243
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.221
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.281
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.237
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.241
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.222
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.36
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.379
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     0.352
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.354
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.249
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.225
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.246
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.238
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.238
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.25
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.233
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.229
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.244
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.242
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.24
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.243
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.241
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.243
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.259
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.229
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.228
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.237
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.303
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.291
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.321
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.25
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.316
    ]
   }
  },
  "gliner2-large": {
   "display": "GLiNER2 large (Fastino)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 71,
     "w": 1,
     "f": 0,
     "n": 0,
     "accuracy": 0.9861
    },
    "standard": {
     "c": 60,
     "w": 36,
     "f": 0,
     "n": 0,
     "accuracy": 0.625
    },
    "judge": {
     "c": 89,
     "w": 57,
     "f": 0,
     "n": 0,
     "accuracy": 0.6096
    },
    "hard": {
     "c": 80,
     "w": 140,
     "f": 0,
     "n": 0,
     "accuracy": 0.3636
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 17,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9444
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adequacy": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     },
     "routing": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     }
    },
    "judge": {
     "routing": {
      "c": 27,
      "w": 51,
      "f": 0,
      "n": 0,
      "accuracy": 0.3462
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 10,
      "w": 28,
      "f": 0,
      "n": 0,
      "accuracy": 0.2632
     },
     "probability": {
      "c": 5,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "temporal_numeric": {
      "c": 7,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.2333
     },
     "ambiguous": {
      "c": 4,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.2857
     },
     "multi_hop": {
      "c": 9,
      "w": 26,
      "f": 0,
      "n": 0,
      "accuracy": 0.2571
     },
     "tradeoff": {
      "c": 2,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.1667
     },
     "adversarial": {
      "c": 7,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.5833
     },
     "trap": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 7,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.7
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.758
    ],
    "easy-intent-01": [
     "c",
     0.659
    ],
    "easy-intent-02": [
     "c",
     0.692
    ],
    "easy-intent-03": [
     "c",
     0.638
    ],
    "easy-intent-04": [
     "c",
     0.676
    ],
    "easy-intent-05": [
     "c",
     0.744
    ],
    "easy-intent-06": [
     "c",
     0.616
    ],
    "easy-intent-07": [
     "c",
     0.601
    ],
    "easy-intent-08": [
     "c",
     0.651
    ],
    "easy-intent-09": [
     "c",
     0.626
    ],
    "easy-intent-10": [
     "c",
     0.61
    ],
    "easy-intent-11": [
     "c",
     0.615
    ],
    "easy-fact-00": [
     "c",
     0.515
    ],
    "easy-fact-01": [
     "c",
     0.524
    ],
    "easy-fact-02": [
     "c",
     0.527
    ],
    "easy-fact-03": [
     "c",
     0.55
    ],
    "easy-fact-04": [
     "c",
     0.525
    ],
    "easy-fact-05": [
     "c",
     0.524
    ],
    "easy-fact-06": [
     "c",
     0.534
    ],
    "easy-fact-07": [
     "c",
     0.52
    ],
    "easy-fact-08": [
     "c",
     0.49
    ],
    "easy-fact-09": [
     "c",
     0.507
    ],
    "easy-fact-10": [
     "c",
     0.564
    ],
    "easy-fact-11": [
     "c",
     0.594
    ],
    "easy-extraction-00": [
     "c",
     0.586
    ],
    "easy-extraction-01": [
     "c",
     0.538
    ],
    "easy-extraction-02": [
     "c",
     0.544
    ],
    "easy-extraction-03": [
     "c",
     0.508
    ],
    "easy-extraction-04": [
     "c",
     0.498
    ],
    "easy-extraction-05": [
     "c",
     0.505
    ],
    "easy-extraction-06": [
     "c",
     0.492
    ],
    "easy-extraction-07": [
     "c",
     0.486
    ],
    "easy-extraction-08": [
     "c",
     0.48
    ],
    "easy-extraction-09": [
     "c",
     0.501
    ],
    "easy-extraction-10": [
     "c",
     0.577
    ],
    "easy-extraction-11": [
     "c",
     0.601
    ],
    "easy-tool_selection-00": [
     "c",
     0.679
    ],
    "easy-tool_selection-01": [
     "c",
     0.677
    ],
    "easy-tool_selection-02": [
     "c",
     0.653
    ],
    "easy-tool_selection-03": [
     "c",
     0.637
    ],
    "easy-tool_selection-04": [
     "c",
     0.652
    ],
    "easy-tool_selection-05": [
     "c",
     0.649
    ],
    "easy-tool_selection-06": [
     "c",
     0.657
    ],
    "easy-tool_selection-07": [
     "c",
     0.672
    ],
    "easy-tool_selection-08": [
     "c",
     0.701
    ],
    "easy-tool_selection-09": [
     "c",
     0.747
    ],
    "easy-tool_selection-10": [
     "c",
     0.665
    ],
    "easy-tool_selection-11": [
     "c",
     0.659
    ],
    "original-policy-01-0": [
     "c",
     0.722
    ],
    "original-policy-01-1": [
     "c",
     0.627
    ],
    "original-policy-02-0": [
     "w",
     0.552
    ],
    "original-policy-02-1": [
     "w",
     0.555
    ],
    "original-policy-03-0": [
     "w",
     0.556
    ],
    "original-policy-03-1": [
     "w",
     0.551
    ],
    "original-policy-04-0": [
     "c",
     0.562
    ],
    "original-policy-04-1": [
     "c",
     0.556
    ],
    "original-policy-05-0": [
     "c",
     0.569
    ],
    "original-policy-05-1": [
     "c",
     0.55
    ],
    "original-policy-06-0": [
     "w",
     0.559
    ],
    "original-policy-06-1": [
     "w",
     0.539
    ],
    "original-intent-01-0": [
     "c",
     0.579
    ],
    "original-intent-01-1": [
     "w",
     0.555
    ],
    "original-intent-02-0": [
     "w",
     0.563
    ],
    "original-intent-02-1": [
     "c",
     0.578
    ],
    "original-intent-03-0": [
     "w",
     0.561
    ],
    "original-intent-03-1": [
     "c",
     0.576
    ],
    "original-intent-04-0": [
     "c",
     0.591
    ],
    "original-intent-04-1": [
     "c",
     0.577
    ],
    "original-intent-05-0": [
     "w",
     0.571
    ],
    "original-intent-05-1": [
     "c",
     0.564
    ],
    "original-intent-06-0": [
     "c",
     0.551
    ],
    "original-intent-06-1": [
     "c",
     0.599
    ],
    "original-ordinal-01-0": [
     "c",
     0.594
    ],
    "original-ordinal-01-1": [
     "c",
     0.637
    ],
    "original-ordinal-02-0": [
     "c",
     0.595
    ],
    "original-ordinal-02-1": [
     "c",
     0.599
    ],
    "original-ordinal-03-0": [
     "c",
     0.586
    ],
    "original-ordinal-03-1": [
     "c",
     0.582
    ],
    "original-ordinal-04-0": [
     "c",
     0.581
    ],
    "original-ordinal-04-1": [
     "c",
     0.599
    ],
    "original-ordinal-05-0": [
     "c",
     0.647
    ],
    "original-ordinal-05-1": [
     "c",
     0.605
    ],
    "original-ordinal-06-0": [
     "c",
     0.592
    ],
    "original-ordinal-06-1": [
     "c",
     0.676
    ],
    "original-extraction-01-0": [
     "w",
     0.575
    ],
    "original-extraction-01-1": [
     "w",
     0.623
    ],
    "original-extraction-02-0": [
     "c",
     0.566
    ],
    "original-extraction-02-1": [
     "c",
     0.59
    ],
    "original-extraction-03-0": [
     "w",
     0.614
    ],
    "original-extraction-03-1": [
     "w",
     0.544
    ],
    "original-extraction-04-0": [
     "w",
     0.661
    ],
    "original-extraction-04-1": [
     "w",
     0.54
    ],
    "original-extraction-05-0": [
     "c",
     0.574
    ],
    "original-extraction-05-1": [
     "c",
     0.614
    ],
    "original-extraction-06-0": [
     "w",
     0.541
    ],
    "original-extraction-06-1": [
     "w",
     0.53
    ],
    "original-adequacy-01-0": [
     "c",
     0.545
    ],
    "original-adequacy-01-1": [
     "c",
     0.49
    ],
    "original-adequacy-02-0": [
     "w",
     0.517
    ],
    "original-adequacy-02-1": [
     "c",
     0.523
    ],
    "original-adequacy-03-0": [
     "w",
     0.535
    ],
    "original-adequacy-03-1": [
     "w",
     0.532
    ],
    "original-adequacy-04-0": [
     "w",
     0.535
    ],
    "original-adequacy-04-1": [
     "w",
     0.513
    ],
    "original-adequacy-05-0": [
     "w",
     0.507
    ],
    "original-adequacy-05-1": [
     "w",
     0.511
    ],
    "original-adequacy-06-0": [
     "c",
     0.576
    ],
    "original-adequacy-06-1": [
     "c",
     0.529
    ],
    "original-routing-01-0": [
     "c",
     0.651
    ],
    "original-routing-01-1": [
     "w",
     0.662
    ],
    "original-routing-02-0": [
     "c",
     0.671
    ],
    "original-routing-02-1": [
     "c",
     0.668
    ],
    "original-routing-03-0": [
     "c",
     0.753
    ],
    "original-routing-03-1": [
     "c",
     0.721
    ],
    "original-routing-04-0": [
     "c",
     0.683
    ],
    "original-routing-04-1": [
     "c",
     0.662
    ],
    "original-routing-05-0": [
     "w",
     0.695
    ],
    "original-routing-05-1": [
     "w",
     0.674
    ],
    "original-routing-06-0": [
     "w",
     0.686
    ],
    "original-routing-06-1": [
     "w",
     0.705
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     114.85
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     41.82
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     29.116
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     43.864
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     27.872
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     27.534
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     38.175
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     24.825
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.698
    ],
    "hard-opus-a-probability-04": [
     "w",
     5.128
    ],
    "hard-opus-a-probability-07": [
     "w",
     5.067
    ],
    "hard-opus-a-probability-08": [
     "w",
     1.263
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     1.665
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     1.696
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     7.121
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     2.139
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     2.193
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     2.108
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     1.898
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     1.921
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     2.704
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     2.648
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     1.778
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     3.251
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     63.813
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     74.616
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     41.222
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     76.848
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     65.218
    ],
    "hard-opus-b-probability-01": [
     "w",
     17.566
    ],
    "hard-opus-b-probability-02": [
     "w",
     2.166
    ],
    "hard-opus-b-probability-03": [
     "c",
     5.618
    ],
    "hard-opus-b-probability-04": [
     "w",
     2.336
    ],
    "hard-opus-b-probability-06": [
     "c",
     5.262
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     6.252
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     2.997
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     3.262
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     3.254
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     2.264
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     2.429
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     58.678
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     64.244
    ],
    "hard-opus-c-long_policy-04": [
     "w",
     47.863
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     74.797
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     80.791
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     55.648
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     48.205
    ],
    "hard-opus-c-probability-03": [
     "w",
     2.419
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     3.522
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     3.427
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     3.922
    ],
    "hard-opus-c-temporal_numeric-06": [
     "c",
     5.25
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     3.6
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     4.044
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     1.371
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     1.064
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     1.166
    ],
    "hard-sol-a-adversarial-08": [
     "w",
     0.922
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     1.003
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     1.191
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     40.134
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     52.051
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     41.867
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     43.31
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     42.561
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     45.244
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     44.902
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     46.987
    ],
    "hard-sol-a-trap-02": [
     "w",
     1.62
    ],
    "hard-sol-a-trap-04": [
     "c",
     1.236
    ],
    "hard-sol-a-trap-05": [
     "c",
     1.843
    ],
    "hard-sol-a-trap-06": [
     "c",
     1.787
    ],
    "hard-sol-a-trap-08": [
     "w",
     1.103
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.015
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.903
    ],
    "hard-sol-a-trap-15": [
     "c",
     1.316
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.73
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     1.242
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     1.202
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.876
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.974
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     1.013
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.833
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.795
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     40.201
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     37.928
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     39.01
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     47.819
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.593
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.58
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     1.539
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     1.34
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     1.466
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.328
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     1.379
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     1.22
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     1.131
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     2.246
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     2.248
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     1.533
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     1.599
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     1.401
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     1.642
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     1.682
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     1.644
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.325
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     39.023
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     33.074
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     35.742
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     2.826
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     38.941
    ]
   }
  },
  "gliner2.5-multi": {
   "display": "GLiNER2.5 multi (Fastino, 287M)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 65,
     "w": 7,
     "f": 0,
     "n": 0,
     "accuracy": 0.9028
    },
    "standard": {
     "c": 49,
     "w": 47,
     "f": 0,
     "n": 0,
     "accuracy": 0.5104
    },
    "judge": {
     "c": 64,
     "w": 82,
     "f": 0,
     "n": 0,
     "accuracy": 0.4384
    },
    "hard": {
     "c": 83,
     "w": 137,
     "f": 0,
     "n": 0,
     "accuracy": 0.3773
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 16,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8889
     },
     "fact": {
      "c": 13,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.7222
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "ordinal": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "extraction": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "adequacy": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     },
     "routing": {
      "c": 4,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     }
    },
    "judge": {
     "routing": {
      "c": 13,
      "w": 65,
      "f": 0,
      "n": 0,
      "accuracy": 0.1667
     },
     "adequacy": {
      "c": 51,
      "w": 17,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     }
    },
    "hard": {
     "long_policy": {
      "c": 14,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.3684
     },
     "probability": {
      "c": 9,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.45
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 6,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.4286
     },
     "multi_hop": {
      "c": 13,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.3714
     },
     "tradeoff": {
      "c": 4,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "adversarial": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "trap": {
      "c": 5,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.3125
     },
     "judge_hard": {
      "c": 17,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.5152
     },
     "routing_hard": {
      "c": 1,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.1
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.319
    ],
    "easy-intent-01": [
     "c",
     0.314
    ],
    "easy-intent-02": [
     "c",
     0.235
    ],
    "easy-intent-03": [
     "c",
     0.258
    ],
    "easy-intent-04": [
     "c",
     0.276
    ],
    "easy-intent-05": [
     "c",
     0.26
    ],
    "easy-intent-06": [
     "w",
     0.249
    ],
    "easy-intent-07": [
     "c",
     0.204
    ],
    "easy-intent-08": [
     "c",
     0.196
    ],
    "easy-intent-09": [
     "c",
     0.214
    ],
    "easy-intent-10": [
     "c",
     0.195
    ],
    "easy-intent-11": [
     "c",
     0.195
    ],
    "easy-fact-00": [
     "c",
     0.174
    ],
    "easy-fact-01": [
     "w",
     0.164
    ],
    "easy-fact-02": [
     "c",
     0.177
    ],
    "easy-fact-03": [
     "w",
     0.172
    ],
    "easy-fact-04": [
     "c",
     0.179
    ],
    "easy-fact-05": [
     "c",
     0.173
    ],
    "easy-fact-06": [
     "c",
     0.188
    ],
    "easy-fact-07": [
     "c",
     0.172
    ],
    "easy-fact-08": [
     "c",
     0.169
    ],
    "easy-fact-09": [
     "w",
     0.175
    ],
    "easy-fact-10": [
     "c",
     0.177
    ],
    "easy-fact-11": [
     "c",
     0.233
    ],
    "easy-extraction-00": [
     "c",
     0.187
    ],
    "easy-extraction-01": [
     "c",
     0.188
    ],
    "easy-extraction-02": [
     "c",
     0.169
    ],
    "easy-extraction-03": [
     "c",
     0.164
    ],
    "easy-extraction-04": [
     "c",
     0.169
    ],
    "easy-extraction-05": [
     "c",
     0.157
    ],
    "easy-extraction-06": [
     "c",
     0.155
    ],
    "easy-extraction-07": [
     "c",
     0.151
    ],
    "easy-extraction-08": [
     "c",
     0.164
    ],
    "easy-extraction-09": [
     "c",
     0.169
    ],
    "easy-extraction-10": [
     "c",
     0.182
    ],
    "easy-extraction-11": [
     "c",
     0.167
    ],
    "easy-tool_selection-00": [
     "c",
     0.219
    ],
    "easy-tool_selection-01": [
     "c",
     0.215
    ],
    "easy-tool_selection-02": [
     "c",
     0.24
    ],
    "easy-tool_selection-03": [
     "c",
     0.21
    ],
    "easy-tool_selection-04": [
     "c",
     0.238
    ],
    "easy-tool_selection-05": [
     "c",
     0.269
    ],
    "easy-tool_selection-06": [
     "c",
     0.252
    ],
    "easy-tool_selection-07": [
     "c",
     0.381
    ],
    "easy-tool_selection-08": [
     "c",
     0.215
    ],
    "easy-tool_selection-09": [
     "c",
     0.235
    ],
    "easy-tool_selection-10": [
     "c",
     0.225
    ],
    "easy-tool_selection-11": [
     "c",
     0.238
    ],
    "original-policy-01-0": [
     "c",
     0.28
    ],
    "original-policy-01-1": [
     "w",
     0.205
    ],
    "original-policy-02-0": [
     "c",
     0.248
    ],
    "original-policy-02-1": [
     "c",
     0.216
    ],
    "original-policy-03-0": [
     "w",
     0.208
    ],
    "original-policy-03-1": [
     "w",
     0.188
    ],
    "original-policy-04-0": [
     "w",
     0.19
    ],
    "original-policy-04-1": [
     "w",
     0.194
    ],
    "original-policy-05-0": [
     "w",
     0.194
    ],
    "original-policy-05-1": [
     "c",
     0.192
    ],
    "original-policy-06-0": [
     "c",
     0.2
    ],
    "original-policy-06-1": [
     "c",
     0.201
    ],
    "original-intent-01-0": [
     "w",
     0.237
    ],
    "original-intent-01-1": [
     "w",
     0.316
    ],
    "original-intent-02-0": [
     "w",
     0.227
    ],
    "original-intent-02-1": [
     "w",
     0.217
    ],
    "original-intent-03-0": [
     "c",
     0.251
    ],
    "original-intent-03-1": [
     "c",
     0.229
    ],
    "original-intent-04-0": [
     "c",
     0.224
    ],
    "original-intent-04-1": [
     "w",
     0.206
    ],
    "original-intent-05-0": [
     "c",
     0.216
    ],
    "original-intent-05-1": [
     "c",
     0.216
    ],
    "original-intent-06-0": [
     "w",
     0.273
    ],
    "original-intent-06-1": [
     "c",
     0.29
    ],
    "original-ordinal-01-0": [
     "c",
     0.299
    ],
    "original-ordinal-01-1": [
     "c",
     0.274
    ],
    "original-ordinal-02-0": [
     "c",
     0.281
    ],
    "original-ordinal-02-1": [
     "c",
     0.243
    ],
    "original-ordinal-03-0": [
     "w",
     0.212
    ],
    "original-ordinal-03-1": [
     "w",
     0.218
    ],
    "original-ordinal-04-0": [
     "c",
     0.211
    ],
    "original-ordinal-04-1": [
     "w",
     0.204
    ],
    "original-ordinal-05-0": [
     "w",
     0.255
    ],
    "original-ordinal-05-1": [
     "c",
     0.209
    ],
    "original-ordinal-06-0": [
     "c",
     0.211
    ],
    "original-ordinal-06-1": [
     "c",
     0.225
    ],
    "original-extraction-01-0": [
     "w",
     0.213
    ],
    "original-extraction-01-1": [
     "w",
     0.188
    ],
    "original-extraction-02-0": [
     "c",
     0.201
    ],
    "original-extraction-02-1": [
     "c",
     0.218
    ],
    "original-extraction-03-0": [
     "w",
     0.224
    ],
    "original-extraction-03-1": [
     "w",
     0.23
    ],
    "original-extraction-04-0": [
     "c",
     0.251
    ],
    "original-extraction-04-1": [
     "w",
     0.402
    ],
    "original-extraction-05-0": [
     "c",
     0.192
    ],
    "original-extraction-05-1": [
     "c",
     0.185
    ],
    "original-extraction-06-0": [
     "c",
     0.194
    ],
    "original-extraction-06-1": [
     "w",
     0.247
    ],
    "original-adequacy-01-0": [
     "w",
     0.261
    ],
    "original-adequacy-01-1": [
     "c",
     0.206
    ],
    "original-adequacy-02-0": [
     "w",
     0.267
    ],
    "original-adequacy-02-1": [
     "c",
     0.269
    ],
    "original-adequacy-03-0": [
     "w",
     0.213
    ],
    "original-adequacy-03-1": [
     "w",
     0.189
    ],
    "original-adequacy-04-0": [
     "w",
     0.19
    ],
    "original-adequacy-04-1": [
     "w",
     0.185
    ],
    "original-adequacy-05-0": [
     "w",
     0.167
    ],
    "original-adequacy-05-1": [
     "w",
     0.168
    ],
    "original-adequacy-06-0": [
     "c",
     0.189
    ],
    "original-adequacy-06-1": [
     "c",
     0.187
    ],
    "original-routing-01-0": [
     "w",
     0.223
    ],
    "original-routing-01-1": [
     "w",
     0.22
    ],
    "original-routing-02-0": [
     "w",
     0.226
    ],
    "original-routing-02-1": [
     "w",
     0.218
    ],
    "original-routing-03-0": [
     "c",
     0.232
    ],
    "original-routing-03-1": [
     "c",
     0.231
    ],
    "original-routing-04-0": [
     "w",
     0.223
    ],
    "original-routing-04-1": [
     "w",
     0.22
    ],
    "original-routing-05-0": [
     "w",
     0.225
    ],
    "original-routing-05-1": [
     "w",
     0.225
    ],
    "original-routing-06-0": [
     "w",
     0.226
    ],
    "original-routing-06-1": [
     "w",
     0.223
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     104.27
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     25.796
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     19.813
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     22.109
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     17.683
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     21.714
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     19.371
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     17.454
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.748
    ],
    "hard-opus-a-probability-04": [
     "w",
     1.762
    ],
    "hard-opus-a-probability-07": [
     "w",
     2.216
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.551
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.906
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     1.126
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     3.731
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.876
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.908
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     1.099
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.856
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.936
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     1.284
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     1.32
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     0.88
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     1.113
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     29.833
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     54.876
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     23.269
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     51.203
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     58.657
    ],
    "hard-opus-b-probability-01": [
     "w",
     9.356
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.483
    ],
    "hard-opus-b-probability-03": [
     "c",
     3.083
    ],
    "hard-opus-b-probability-04": [
     "w",
     1.289
    ],
    "hard-opus-b-probability-06": [
     "c",
     3.118
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     3.342
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     2.05
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     1.877
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     2.256
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     1.147
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     1.318
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     36.792
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     39.101
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     32.3
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     55.444
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     67.578
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     35.62
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     32.576
    ],
    "hard-opus-c-probability-03": [
     "w",
     1.301
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     2.485
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     2.366
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.625
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     3.415
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     2.665
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     2.077
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.563
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.45
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.513
    ],
    "hard-sol-a-adversarial-08": [
     "w",
     0.443
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.443
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.574
    ],
    "hard-sol-a-multi_hop-01": [
     "w",
     28.06
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     28.902
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     27.422
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     25.95
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     28.033
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     28.373
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     26.912
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     25.637
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.925
    ],
    "hard-sol-a-trap-04": [
     "w",
     0.864
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.751
    ],
    "hard-sol-a-trap-06": [
     "w",
     1.001
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.777
    ],
    "hard-sol-a-trap-10": [
     "w",
     0.472
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.416
    ],
    "hard-sol-a-trap-15": [
     "w",
     0.602
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.442
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.391
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.538
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.382
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.524
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.6
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.451
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.428
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     32.76
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     33.564
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     30.317
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     29.755
    ],
    "hard-sol-b-routing_hard-01": [
     "w",
     1.277
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     0.669
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     0.786
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.652
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     0.725
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.549
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.524
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.403
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.433
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     1.018
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     0.794
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.698
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.684
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.7
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.883
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.827
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.62
    ],
    "hard-sol-c-judge_hard-15": [
     "w",
     0.676
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     25.373
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     23.753
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     20.804
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.972
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     23.478
    ]
   }
  },
  "gliner2.5-small": {
   "display": "GLiNER2.5 small (Fastino, 74M)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 60,
     "w": 12,
     "f": 0,
     "n": 0,
     "accuracy": 0.8333
    },
    "standard": {
     "c": 46,
     "w": 50,
     "f": 0,
     "n": 0,
     "accuracy": 0.4792
    },
    "judge": {
     "c": 73,
     "w": 73,
     "f": 0,
     "n": 0,
     "accuracy": 0.5
    },
    "hard": {
     "c": 73,
     "w": 147,
     "f": 0,
     "n": 0,
     "accuracy": 0.3318
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 13,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.7222
     },
     "fact": {
      "c": 13,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.7222
     },
     "extraction": {
      "c": 16,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8889
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "ordinal": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "extraction": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adequacy": {
      "c": 4,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "routing": {
      "c": 5,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.3125
     }
    },
    "judge": {
     "routing": {
      "c": 12,
      "w": 66,
      "f": 0,
      "n": 0,
      "accuracy": 0.1538
     },
     "adequacy": {
      "c": 61,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8971
     }
    },
    "hard": {
     "long_policy": {
      "c": 12,
      "w": 26,
      "f": 0,
      "n": 0,
      "accuracy": 0.3158
     },
     "probability": {
      "c": 7,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.35
     },
     "temporal_numeric": {
      "c": 6,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     },
     "ambiguous": {
      "c": 2,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.1429
     },
     "multi_hop": {
      "c": 10,
      "w": 25,
      "f": 0,
      "n": 0,
      "accuracy": 0.2857
     },
     "tradeoff": {
      "c": 7,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.5833
     },
     "adversarial": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "trap": {
      "c": 4,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 2,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "w",
     0.13
    ],
    "easy-intent-01": [
     "c",
     0.093
    ],
    "easy-intent-02": [
     "c",
     0.074
    ],
    "easy-intent-03": [
     "c",
     0.097
    ],
    "easy-intent-04": [
     "c",
     0.072
    ],
    "easy-intent-05": [
     "c",
     0.066
    ],
    "easy-intent-06": [
     "w",
     0.072
    ],
    "easy-intent-07": [
     "c",
     0.07
    ],
    "easy-intent-08": [
     "w",
     0.067
    ],
    "easy-intent-09": [
     "c",
     0.07
    ],
    "easy-intent-10": [
     "c",
     0.069
    ],
    "easy-intent-11": [
     "c",
     0.096
    ],
    "easy-fact-00": [
     "c",
     0.072
    ],
    "easy-fact-01": [
     "w",
     0.053
    ],
    "easy-fact-02": [
     "c",
     0.071
    ],
    "easy-fact-03": [
     "w",
     0.087
    ],
    "easy-fact-04": [
     "w",
     0.117
    ],
    "easy-fact-05": [
     "c",
     0.07
    ],
    "easy-fact-06": [
     "c",
     0.062
    ],
    "easy-fact-07": [
     "c",
     0.083
    ],
    "easy-fact-08": [
     "c",
     0.241
    ],
    "easy-fact-09": [
     "c",
     0.07
    ],
    "easy-fact-10": [
     "c",
     0.075
    ],
    "easy-fact-11": [
     "c",
     0.06
    ],
    "easy-extraction-00": [
     "c",
     0.056
    ],
    "easy-extraction-01": [
     "c",
     0.052
    ],
    "easy-extraction-02": [
     "c",
     0.057
    ],
    "easy-extraction-03": [
     "c",
     0.055
    ],
    "easy-extraction-04": [
     "c",
     0.055
    ],
    "easy-extraction-05": [
     "c",
     0.053
    ],
    "easy-extraction-06": [
     "c",
     0.074
    ],
    "easy-extraction-07": [
     "c",
     0.064
    ],
    "easy-extraction-08": [
     "c",
     0.056
    ],
    "easy-extraction-09": [
     "c",
     0.051
    ],
    "easy-extraction-10": [
     "w",
     0.052
    ],
    "easy-extraction-11": [
     "c",
     0.084
    ],
    "easy-tool_selection-00": [
     "c",
     0.07
    ],
    "easy-tool_selection-01": [
     "c",
     0.077
    ],
    "easy-tool_selection-02": [
     "c",
     0.082
    ],
    "easy-tool_selection-03": [
     "c",
     0.079
    ],
    "easy-tool_selection-04": [
     "c",
     0.074
    ],
    "easy-tool_selection-05": [
     "c",
     0.076
    ],
    "easy-tool_selection-06": [
     "c",
     0.091
    ],
    "easy-tool_selection-07": [
     "c",
     0.088
    ],
    "easy-tool_selection-08": [
     "c",
     0.082
    ],
    "easy-tool_selection-09": [
     "c",
     0.07
    ],
    "easy-tool_selection-10": [
     "c",
     0.069
    ],
    "easy-tool_selection-11": [
     "c",
     0.079
    ],
    "original-policy-01-0": [
     "w",
     0.105
    ],
    "original-policy-01-1": [
     "w",
     0.058
    ],
    "original-policy-02-0": [
     "c",
     0.061
    ],
    "original-policy-02-1": [
     "c",
     0.057
    ],
    "original-policy-03-0": [
     "c",
     0.061
    ],
    "original-policy-03-1": [
     "w",
     0.058
    ],
    "original-policy-04-0": [
     "w",
     0.061
    ],
    "original-policy-04-1": [
     "w",
     0.095
    ],
    "original-policy-05-0": [
     "w",
     0.08
    ],
    "original-policy-05-1": [
     "w",
     0.091
    ],
    "original-policy-06-0": [
     "c",
     0.095
    ],
    "original-policy-06-1": [
     "c",
     0.065
    ],
    "original-intent-01-0": [
     "c",
     0.065
    ],
    "original-intent-01-1": [
     "c",
     0.061
    ],
    "original-intent-02-0": [
     "w",
     0.067
    ],
    "original-intent-02-1": [
     "c",
     0.07
    ],
    "original-intent-03-0": [
     "w",
     0.067
    ],
    "original-intent-03-1": [
     "c",
     0.064
    ],
    "original-intent-04-0": [
     "c",
     0.064
    ],
    "original-intent-04-1": [
     "c",
     0.063
    ],
    "original-intent-05-0": [
     "w",
     0.062
    ],
    "original-intent-05-1": [
     "w",
     0.061
    ],
    "original-intent-06-0": [
     "c",
     0.057
    ],
    "original-intent-06-1": [
     "c",
     0.064
    ],
    "original-ordinal-01-0": [
     "w",
     0.066
    ],
    "original-ordinal-01-1": [
     "w",
     0.066
    ],
    "original-ordinal-02-0": [
     "w",
     0.082
    ],
    "original-ordinal-02-1": [
     "w",
     0.066
    ],
    "original-ordinal-03-0": [
     "c",
     0.063
    ],
    "original-ordinal-03-1": [
     "c",
     0.086
    ],
    "original-ordinal-04-0": [
     "c",
     0.081
    ],
    "original-ordinal-04-1": [
     "c",
     0.079
    ],
    "original-ordinal-05-0": [
     "w",
     0.082
    ],
    "original-ordinal-05-1": [
     "w",
     0.061
    ],
    "original-ordinal-06-0": [
     "c",
     0.064
    ],
    "original-ordinal-06-1": [
     "c",
     0.066
    ],
    "original-extraction-01-0": [
     "w",
     0.062
    ],
    "original-extraction-01-1": [
     "w",
     0.057
    ],
    "original-extraction-02-0": [
     "w",
     0.055
    ],
    "original-extraction-02-1": [
     "c",
     0.056
    ],
    "original-extraction-03-0": [
     "w",
     0.062
    ],
    "original-extraction-03-1": [
     "c",
     0.07
    ],
    "original-extraction-04-0": [
     "w",
     0.065
    ],
    "original-extraction-04-1": [
     "c",
     0.074
    ],
    "original-extraction-05-0": [
     "c",
     0.059
    ],
    "original-extraction-05-1": [
     "c",
     0.061
    ],
    "original-extraction-06-0": [
     "c",
     0.082
    ],
    "original-extraction-06-1": [
     "w",
     0.076
    ],
    "original-adequacy-01-0": [
     "c",
     0.085
    ],
    "original-adequacy-01-1": [
     "w",
     0.059
    ],
    "original-adequacy-02-0": [
     "w",
     0.072
    ],
    "original-adequacy-02-1": [
     "w",
     0.072
    ],
    "original-adequacy-03-0": [
     "w",
     0.221
    ],
    "original-adequacy-03-1": [
     "w",
     0.066
    ],
    "original-adequacy-04-0": [
     "w",
     0.056
    ],
    "original-adequacy-04-1": [
     "w",
     0.076
    ],
    "original-adequacy-05-0": [
     "w",
     0.06
    ],
    "original-adequacy-05-1": [
     "w",
     0.062
    ],
    "original-adequacy-06-0": [
     "w",
     0.056
    ],
    "original-adequacy-06-1": [
     "c",
     0.061
    ],
    "original-routing-01-0": [
     "c",
     0.07
    ],
    "original-routing-01-1": [
     "w",
     0.07
    ],
    "original-routing-02-0": [
     "w",
     0.073
    ],
    "original-routing-02-1": [
     "w",
     0.077
    ],
    "original-routing-03-0": [
     "c",
     0.079
    ],
    "original-routing-03-1": [
     "c",
     0.073
    ],
    "original-routing-04-0": [
     "w",
     0.08
    ],
    "original-routing-04-1": [
     "w",
     0.071
    ],
    "original-routing-05-0": [
     "w",
     0.082
    ],
    "original-routing-05-1": [
     "w",
     0.075
    ],
    "original-routing-06-0": [
     "w",
     0.076
    ],
    "original-routing-06-1": [
     "w",
     0.074
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     20.828
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     7.414
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     5.634
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     6.758
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     4.709
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     4.711
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     5.879
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     4.551
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.165
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.536
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.583
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.135
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     0.183
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.184
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.618
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.221
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     0.265
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.276
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.248
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.295
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.351
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.278
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     0.18
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.25
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     8.989
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     12.805
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     8.14
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     13.724
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     13.533
    ],
    "hard-opus-b-probability-01": [
     "c",
     2.528
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.289
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.685
    ],
    "hard-opus-b-probability-04": [
     "w",
     0.304
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.568
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.696
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.4
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.379
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.399
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.261
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     0.286
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     11.308
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     10.517
    ],
    "hard-opus-c-long_policy-04": [
     "w",
     8.993
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     14.031
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     16.052
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     10.478
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     9.354
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.28
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.448
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.438
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.374
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.591
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.487
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.442
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.146
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.115
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.125
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.093
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.101
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.139
    ],
    "hard-sol-a-multi_hop-01": [
     "w",
     7.918
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     7.79
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     8.526
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     7.971
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     7.356
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     7.567
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     8.14
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     8.587
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.194
    ],
    "hard-sol-a-trap-04": [
     "w",
     0.141
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.174
    ],
    "hard-sol-a-trap-06": [
     "w",
     0.214
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.12
    ],
    "hard-sol-a-trap-10": [
     "w",
     0.107
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.09
    ],
    "hard-sol-a-trap-15": [
     "w",
     0.171
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.13
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.088
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.111
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.104
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.123
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.123
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.104
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.107
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     7.99
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     7.117
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     7.305
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     6.979
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.165
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     0.155
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.173
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.173
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     0.147
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.127
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.12
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.111
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.155
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.279
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.198
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.168
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.153
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.168
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.164
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.233
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.143
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.137
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     6.079
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     6.181
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     6.057
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.314
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     6.559
    ]
   }
  },
  "gliner2": {
   "display": "GLiNER2 (Fastino, gliner2.5-base)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 70,
     "w": 2,
     "f": 0,
     "n": 0,
     "accuracy": 0.9722
    },
    "standard": {
     "c": 64,
     "w": 32,
     "f": 0,
     "n": 0,
     "accuracy": 0.6667
    },
    "judge": {
     "c": 67,
     "w": 79,
     "f": 0,
     "n": 0,
     "accuracy": 0.4589
    },
    "hard": {
     "c": 80,
     "w": 140,
     "f": 0,
     "n": 0,
     "accuracy": 0.3636
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 16,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8889
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "intent": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "ordinal": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "extraction": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "adequacy": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "routing": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     }
    },
    "judge": {
     "routing": {
      "c": 10,
      "w": 68,
      "f": 0,
      "n": 0,
      "accuracy": 0.1282
     },
     "adequacy": {
      "c": 57,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.8382
     }
    },
    "hard": {
     "long_policy": {
      "c": 10,
      "w": 28,
      "f": 0,
      "n": 0,
      "accuracy": 0.2632
     },
     "probability": {
      "c": 6,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "temporal_numeric": {
      "c": 6,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     },
     "ambiguous": {
      "c": 5,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.3571
     },
     "multi_hop": {
      "c": 13,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.3714
     },
     "tradeoff": {
      "c": 6,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adversarial": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "trap": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     },
     "judge_hard": {
      "c": 15,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4545
     },
     "routing_hard": {
      "c": 7,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.7
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.327
    ],
    "easy-intent-01": [
     "c",
     0.197
    ],
    "easy-intent-02": [
     "c",
     0.196
    ],
    "easy-intent-03": [
     "c",
     0.214
    ],
    "easy-intent-04": [
     "c",
     0.286
    ],
    "easy-intent-05": [
     "c",
     0.219
    ],
    "easy-intent-06": [
     "c",
     0.212
    ],
    "easy-intent-07": [
     "c",
     0.221
    ],
    "easy-intent-08": [
     "c",
     0.217
    ],
    "easy-intent-09": [
     "c",
     0.216
    ],
    "easy-intent-10": [
     "c",
     0.193
    ],
    "easy-intent-11": [
     "c",
     0.19
    ],
    "easy-fact-00": [
     "c",
     0.179
    ],
    "easy-fact-01": [
     "w",
     0.209
    ],
    "easy-fact-02": [
     "c",
     0.155
    ],
    "easy-fact-03": [
     "c",
     0.157
    ],
    "easy-fact-04": [
     "c",
     0.147
    ],
    "easy-fact-05": [
     "c",
     0.154
    ],
    "easy-fact-06": [
     "c",
     0.169
    ],
    "easy-fact-07": [
     "c",
     0.183
    ],
    "easy-fact-08": [
     "c",
     0.16
    ],
    "easy-fact-09": [
     "c",
     0.18
    ],
    "easy-fact-10": [
     "c",
     0.188
    ],
    "easy-fact-11": [
     "c",
     0.198
    ],
    "easy-extraction-00": [
     "c",
     0.201
    ],
    "easy-extraction-01": [
     "c",
     0.175
    ],
    "easy-extraction-02": [
     "c",
     0.187
    ],
    "easy-extraction-03": [
     "c",
     0.166
    ],
    "easy-extraction-04": [
     "c",
     0.159
    ],
    "easy-extraction-05": [
     "c",
     0.162
    ],
    "easy-extraction-06": [
     "c",
     0.166
    ],
    "easy-extraction-07": [
     "c",
     0.139
    ],
    "easy-extraction-08": [
     "c",
     0.14
    ],
    "easy-extraction-09": [
     "c",
     0.158
    ],
    "easy-extraction-10": [
     "c",
     0.147
    ],
    "easy-extraction-11": [
     "c",
     0.158
    ],
    "easy-tool_selection-00": [
     "c",
     0.268
    ],
    "easy-tool_selection-01": [
     "c",
     0.196
    ],
    "easy-tool_selection-02": [
     "c",
     0.203
    ],
    "easy-tool_selection-03": [
     "c",
     0.201
    ],
    "easy-tool_selection-04": [
     "c",
     0.215
    ],
    "easy-tool_selection-05": [
     "c",
     0.192
    ],
    "easy-tool_selection-06": [
     "c",
     0.192
    ],
    "easy-tool_selection-07": [
     "c",
     0.191
    ],
    "easy-tool_selection-08": [
     "c",
     0.188
    ],
    "easy-tool_selection-09": [
     "c",
     0.197
    ],
    "easy-tool_selection-10": [
     "c",
     0.278
    ],
    "easy-tool_selection-11": [
     "c",
     0.219
    ],
    "original-policy-01-0": [
     "c",
     0.287
    ],
    "original-policy-01-1": [
     "c",
     0.251
    ],
    "original-policy-02-0": [
     "c",
     0.208
    ],
    "original-policy-02-1": [
     "w",
     0.159
    ],
    "original-policy-03-0": [
     "w",
     0.16
    ],
    "original-policy-03-1": [
     "w",
     0.162
    ],
    "original-policy-04-0": [
     "c",
     0.167
    ],
    "original-policy-04-1": [
     "w",
     0.163
    ],
    "original-policy-05-0": [
     "c",
     0.175
    ],
    "original-policy-05-1": [
     "c",
     0.164
    ],
    "original-policy-06-0": [
     "w",
     0.165
    ],
    "original-policy-06-1": [
     "c",
     0.162
    ],
    "original-intent-01-0": [
     "c",
     0.182
    ],
    "original-intent-01-1": [
     "c",
     0.163
    ],
    "original-intent-02-0": [
     "w",
     0.171
    ],
    "original-intent-02-1": [
     "c",
     0.168
    ],
    "original-intent-03-0": [
     "w",
     0.167
    ],
    "original-intent-03-1": [
     "c",
     0.163
    ],
    "original-intent-04-0": [
     "c",
     0.183
    ],
    "original-intent-04-1": [
     "c",
     0.176
    ],
    "original-intent-05-0": [
     "c",
     0.221
    ],
    "original-intent-05-1": [
     "c",
     0.167
    ],
    "original-intent-06-0": [
     "c",
     0.171
    ],
    "original-intent-06-1": [
     "c",
     0.193
    ],
    "original-ordinal-01-0": [
     "c",
     0.167
    ],
    "original-ordinal-01-1": [
     "c",
     0.163
    ],
    "original-ordinal-02-0": [
     "c",
     0.171
    ],
    "original-ordinal-02-1": [
     "c",
     0.172
    ],
    "original-ordinal-03-0": [
     "c",
     0.167
    ],
    "original-ordinal-03-1": [
     "c",
     0.169
    ],
    "original-ordinal-04-0": [
     "c",
     0.167
    ],
    "original-ordinal-04-1": [
     "c",
     0.167
    ],
    "original-ordinal-05-0": [
     "c",
     0.167
    ],
    "original-ordinal-05-1": [
     "c",
     0.177
    ],
    "original-ordinal-06-0": [
     "c",
     0.195
    ],
    "original-ordinal-06-1": [
     "c",
     0.181
    ],
    "original-extraction-01-0": [
     "w",
     0.154
    ],
    "original-extraction-01-1": [
     "w",
     0.163
    ],
    "original-extraction-02-0": [
     "c",
     0.162
    ],
    "original-extraction-02-1": [
     "c",
     0.153
    ],
    "original-extraction-03-0": [
     "w",
     0.159
    ],
    "original-extraction-03-1": [
     "w",
     0.154
    ],
    "original-extraction-04-0": [
     "c",
     0.155
    ],
    "original-extraction-04-1": [
     "w",
     0.152
    ],
    "original-extraction-05-0": [
     "c",
     0.153
    ],
    "original-extraction-05-1": [
     "c",
     0.148
    ],
    "original-extraction-06-0": [
     "c",
     0.153
    ],
    "original-extraction-06-1": [
     "w",
     0.155
    ],
    "original-adequacy-01-0": [
     "w",
     0.155
    ],
    "original-adequacy-01-1": [
     "w",
     0.173
    ],
    "original-adequacy-02-0": [
     "c",
     0.215
    ],
    "original-adequacy-02-1": [
     "c",
     0.2
    ],
    "original-adequacy-03-0": [
     "c",
     0.194
    ],
    "original-adequacy-03-1": [
     "c",
     0.215
    ],
    "original-adequacy-04-0": [
     "w",
     0.203
    ],
    "original-adequacy-04-1": [
     "w",
     0.321
    ],
    "original-adequacy-05-0": [
     "c",
     0.162
    ],
    "original-adequacy-05-1": [
     "c",
     0.21
    ],
    "original-adequacy-06-0": [
     "w",
     0.175
    ],
    "original-adequacy-06-1": [
     "w",
     0.19
    ],
    "original-routing-01-0": [
     "c",
     0.206
    ],
    "original-routing-01-1": [
     "w",
     0.189
    ],
    "original-routing-02-0": [
     "c",
     0.197
    ],
    "original-routing-02-1": [
     "w",
     0.19
    ],
    "original-routing-03-0": [
     "c",
     0.201
    ],
    "original-routing-03-1": [
     "c",
     0.202
    ],
    "original-routing-04-0": [
     "c",
     0.225
    ],
    "original-routing-04-1": [
     "w",
     0.227
    ],
    "original-routing-05-0": [
     "w",
     0.196
    ],
    "original-routing-05-1": [
     "w",
     0.196
    ],
    "original-routing-06-0": [
     "w",
     0.192
    ],
    "original-routing-06-1": [
     "w",
     0.197
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     82.344
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     16.416
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     10.726
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     12.309
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     9.262
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     10.956
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     10.739
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     9.654
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.176
    ],
    "hard-opus-a-probability-04": [
     "c",
     1.854
    ],
    "hard-opus-a-probability-07": [
     "w",
     2.167
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.581
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.598
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.75
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     2.693
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.783
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     0.745
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.876
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.644
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.64
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.9
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.988
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.567
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.864
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     17.67
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     24.351
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     12.41
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     27.616
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     35.574
    ],
    "hard-opus-b-probability-01": [
     "w",
     7.283
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.107
    ],
    "hard-opus-b-probability-03": [
     "w",
     2.981
    ],
    "hard-opus-b-probability-04": [
     "w",
     0.819
    ],
    "hard-opus-b-probability-06": [
     "c",
     2.068
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     1.98
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     1.16
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     1.214
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     1.497
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.784
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     1.106
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     26.512
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     27.931
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     16.827
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     27.611
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     87.643
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     24.191
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     18.789
    ],
    "hard-opus-c-probability-03": [
     "w",
     1.334
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     1.683
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     1.461
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.259
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     1.735
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     1.431
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     1.303
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.579
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.415
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     0.431
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.403
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.38
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.545
    ],
    "hard-sol-a-multi_hop-01": [
     "w",
     17.213
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     18.116
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     15.761
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     18.962
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     16.212
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     15.515
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     13.686
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     15.388
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.395
    ],
    "hard-sol-a-trap-04": [
     "w",
     0.316
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.528
    ],
    "hard-sol-a-trap-06": [
     "w",
     0.477
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.384
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.359
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.369
    ],
    "hard-sol-a-trap-15": [
     "w",
     0.426
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.358
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.333
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.393
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.282
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.286
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.343
    ],
    "hard-sol-b-judge_hard-15": [
     "w",
     0.258
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.252
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     16.174
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     14.187
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     13.748
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     14.744
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.536
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.46
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.437
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.405
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.415
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.376
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.435
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.321
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.367
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.584
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     0.41
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.412
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.489
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.438
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.421
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.475
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.534
    ],
    "hard-sol-c-judge_hard-15": [
     "w",
     0.491
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     13.085
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     11.409
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     11.936
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.732
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     13.095
    ]
   }
  },
  "jeff": {
   "display": "jeff (Logan Markewich, GLiFormer 400M)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 73,
     "w": 23,
     "f": 0,
     "n": 0,
     "accuracy": 0.7604
    },
    "judge": {
     "c": 90,
     "w": 56,
     "f": 0,
     "n": 0,
     "accuracy": 0.6164
    },
    "hard": {
     "c": 83,
     "w": 137,
     "f": 0,
     "n": 0,
     "accuracy": 0.3773
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "intent": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     },
     "ordinal": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 7,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.4375
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 38,
      "w": 40,
      "f": 0,
      "n": 0,
      "accuracy": 0.4872
     },
     "adequacy": {
      "c": 52,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.7647
     }
    },
    "hard": {
     "long_policy": {
      "c": 5,
      "w": 33,
      "f": 0,
      "n": 0,
      "accuracy": 0.1316
     },
     "probability": {
      "c": 7,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.35
     },
     "temporal_numeric": {
      "c": 10,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "ambiguous": {
      "c": 0,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.0
     },
     "multi_hop": {
      "c": 17,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4857
     },
     "tradeoff": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "adversarial": {
      "c": 7,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.5833
     },
     "trap": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "judge_hard": {
      "c": 17,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.5152
     },
     "routing_hard": {
      "c": 4,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.636
    ],
    "easy-intent-01": [
     "c",
     0.612
    ],
    "easy-intent-02": [
     "c",
     0.628
    ],
    "easy-intent-03": [
     "c",
     0.61
    ],
    "easy-intent-04": [
     "c",
     0.608
    ],
    "easy-intent-05": [
     "c",
     0.618
    ],
    "easy-intent-06": [
     "c",
     0.588
    ],
    "easy-intent-07": [
     "c",
     0.558
    ],
    "easy-intent-08": [
     "c",
     0.637
    ],
    "easy-intent-09": [
     "c",
     0.581
    ],
    "easy-intent-10": [
     "c",
     0.572
    ],
    "easy-intent-11": [
     "c",
     0.571
    ],
    "easy-fact-00": [
     "c",
     0.52
    ],
    "easy-fact-01": [
     "c",
     0.498
    ],
    "easy-fact-02": [
     "c",
     0.512
    ],
    "easy-fact-03": [
     "c",
     0.696
    ],
    "easy-fact-04": [
     "c",
     0.829
    ],
    "easy-fact-05": [
     "c",
     0.575
    ],
    "easy-fact-06": [
     "c",
     0.57
    ],
    "easy-fact-07": [
     "c",
     0.528
    ],
    "easy-fact-08": [
     "c",
     0.491
    ],
    "easy-fact-09": [
     "c",
     0.55
    ],
    "easy-fact-10": [
     "c",
     0.528
    ],
    "easy-fact-11": [
     "c",
     0.55
    ],
    "easy-extraction-00": [
     "c",
     0.487
    ],
    "easy-extraction-01": [
     "c",
     0.515
    ],
    "easy-extraction-02": [
     "c",
     0.57
    ],
    "easy-extraction-03": [
     "c",
     0.509
    ],
    "easy-extraction-04": [
     "c",
     0.482
    ],
    "easy-extraction-05": [
     "c",
     0.483
    ],
    "easy-extraction-06": [
     "c",
     0.487
    ],
    "easy-extraction-07": [
     "c",
     0.481
    ],
    "easy-extraction-08": [
     "c",
     0.491
    ],
    "easy-extraction-09": [
     "c",
     0.535
    ],
    "easy-extraction-10": [
     "c",
     0.483
    ],
    "easy-extraction-11": [
     "c",
     0.496
    ],
    "easy-tool_selection-00": [
     "c",
     0.62
    ],
    "easy-tool_selection-01": [
     "c",
     0.602
    ],
    "easy-tool_selection-02": [
     "c",
     0.599
    ],
    "easy-tool_selection-03": [
     "c",
     0.601
    ],
    "easy-tool_selection-04": [
     "c",
     0.608
    ],
    "easy-tool_selection-05": [
     "c",
     0.599
    ],
    "easy-tool_selection-06": [
     "c",
     0.63
    ],
    "easy-tool_selection-07": [
     "c",
     0.629
    ],
    "easy-tool_selection-08": [
     "c",
     0.619
    ],
    "easy-tool_selection-09": [
     "c",
     0.796
    ],
    "easy-tool_selection-10": [
     "c",
     0.982
    ],
    "easy-tool_selection-11": [
     "c",
     1.006
    ],
    "original-policy-01-0": [
     "c",
     0.644
    ],
    "original-policy-01-1": [
     "c",
     0.568
    ],
    "original-policy-02-0": [
     "c",
     0.553
    ],
    "original-policy-02-1": [
     "c",
     0.535
    ],
    "original-policy-03-0": [
     "c",
     0.548
    ],
    "original-policy-03-1": [
     "c",
     0.566
    ],
    "original-policy-04-0": [
     "c",
     0.605
    ],
    "original-policy-04-1": [
     "w",
     0.67
    ],
    "original-policy-05-0": [
     "c",
     0.791
    ],
    "original-policy-05-1": [
     "c",
     0.667
    ],
    "original-policy-06-0": [
     "c",
     0.632
    ],
    "original-policy-06-1": [
     "c",
     0.833
    ],
    "original-intent-01-0": [
     "c",
     0.764
    ],
    "original-intent-01-1": [
     "w",
     0.618
    ],
    "original-intent-02-0": [
     "c",
     0.591
    ],
    "original-intent-02-1": [
     "w",
     0.595
    ],
    "original-intent-03-0": [
     "w",
     0.556
    ],
    "original-intent-03-1": [
     "c",
     0.584
    ],
    "original-intent-04-0": [
     "c",
     0.719
    ],
    "original-intent-04-1": [
     "c",
     0.564
    ],
    "original-intent-05-0": [
     "w",
     0.531
    ],
    "original-intent-05-1": [
     "w",
     0.545
    ],
    "original-intent-06-0": [
     "w",
     0.585
    ],
    "original-intent-06-1": [
     "w",
     0.584
    ],
    "original-ordinal-01-0": [
     "c",
     0.522
    ],
    "original-ordinal-01-1": [
     "c",
     0.55
    ],
    "original-ordinal-02-0": [
     "c",
     0.564
    ],
    "original-ordinal-02-1": [
     "c",
     0.591
    ],
    "original-ordinal-03-0": [
     "w",
     0.526
    ],
    "original-ordinal-03-1": [
     "w",
     0.6
    ],
    "original-ordinal-04-0": [
     "c",
     0.698
    ],
    "original-ordinal-04-1": [
     "c",
     0.631
    ],
    "original-ordinal-05-0": [
     "c",
     0.637
    ],
    "original-ordinal-05-1": [
     "c",
     0.649
    ],
    "original-ordinal-06-0": [
     "c",
     0.676
    ],
    "original-ordinal-06-1": [
     "c",
     0.69
    ],
    "original-extraction-01-0": [
     "c",
     0.669
    ],
    "original-extraction-01-1": [
     "c",
     0.771
    ],
    "original-extraction-02-0": [
     "c",
     0.938
    ],
    "original-extraction-02-1": [
     "c",
     0.718
    ],
    "original-extraction-03-0": [
     "c",
     0.62
    ],
    "original-extraction-03-1": [
     "c",
     0.982
    ],
    "original-extraction-04-0": [
     "c",
     0.725
    ],
    "original-extraction-04-1": [
     "c",
     0.625
    ],
    "original-extraction-05-0": [
     "c",
     0.796
    ],
    "original-extraction-05-1": [
     "c",
     0.856
    ],
    "original-extraction-06-0": [
     "c",
     0.578
    ],
    "original-extraction-06-1": [
     "c",
     0.582
    ],
    "original-adequacy-01-0": [
     "w",
     0.656
    ],
    "original-adequacy-01-1": [
     "c",
     0.6
    ],
    "original-adequacy-02-0": [
     "w",
     0.584
    ],
    "original-adequacy-02-1": [
     "c",
     0.558
    ],
    "original-adequacy-03-0": [
     "w",
     0.564
    ],
    "original-adequacy-03-1": [
     "w",
     0.675
    ],
    "original-adequacy-04-0": [
     "w",
     0.587
    ],
    "original-adequacy-04-1": [
     "w",
     0.561
    ],
    "original-adequacy-05-0": [
     "w",
     0.558
    ],
    "original-adequacy-05-1": [
     "w",
     0.56
    ],
    "original-adequacy-06-0": [
     "c",
     0.63
    ],
    "original-adequacy-06-1": [
     "c",
     0.563
    ],
    "original-routing-01-0": [
     "c",
     0.854
    ],
    "original-routing-01-1": [
     "c",
     0.786
    ],
    "original-routing-02-0": [
     "c",
     0.686
    ],
    "original-routing-02-1": [
     "c",
     0.712
    ],
    "original-routing-03-0": [
     "c",
     0.711
    ],
    "original-routing-03-1": [
     "c",
     0.726
    ],
    "original-routing-04-0": [
     "c",
     0.666
    ],
    "original-routing-04-1": [
     "c",
     0.784
    ],
    "original-routing-05-0": [
     "c",
     0.822
    ],
    "original-routing-05-1": [
     "c",
     0.805
    ],
    "original-routing-06-0": [
     "c",
     0.754
    ],
    "original-routing-06-1": [
     "c",
     0.672
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     97.424
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     34.215
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     29.723
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     33.194
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     26.616
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     28.287
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     32.416
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     26.091
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.932
    ],
    "hard-opus-a-probability-04": [
     "w",
     6.51
    ],
    "hard-opus-a-probability-07": [
     "w",
     6.024
    ],
    "hard-opus-a-probability-08": [
     "c",
     1.472
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     1.858
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     1.74
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     7.431
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     2.001
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     2.069
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     2.113
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     1.811
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     2.016
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     2.458
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     2.321
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     1.677
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     2.257
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     41.15
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     63.625
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     33.354
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     64.37
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     64.613
    ],
    "hard-opus-b-probability-01": [
     "w",
     17.655
    ],
    "hard-opus-b-probability-02": [
     "w",
     2.382
    ],
    "hard-opus-b-probability-03": [
     "c",
     7.293
    ],
    "hard-opus-b-probability-04": [
     "w",
     2.177
    ],
    "hard-opus-b-probability-06": [
     "w",
     5.219
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     5.083
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     2.784
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     3.109
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     3.828
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     2.37
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     2.456
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     52.024
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     52.414
    ],
    "hard-opus-c-long_policy-04": [
     "w",
     43.054
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     63.127
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     83.479
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     54.744
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     48.308
    ],
    "hard-opus-c-probability-03": [
     "c",
     2.694
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     4.744
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     3.811
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     2.81
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     4.632
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     3.738
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     4.057
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     1.439
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     1.09
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.18
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.957
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.947
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     1.235
    ],
    "hard-sol-a-multi_hop-01": [
     "w",
     41.379
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     41.895
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     39.802
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     36.736
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     37.96
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     39.636
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     40.562
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     38.118
    ],
    "hard-sol-a-trap-02": [
     "c",
     1.224
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.897
    ],
    "hard-sol-a-trap-05": [
     "w",
     1.374
    ],
    "hard-sol-a-trap-06": [
     "w",
     1.46
    ],
    "hard-sol-a-trap-08": [
     "w",
     1.156
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.003
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.945
    ],
    "hard-sol-a-trap-15": [
     "w",
     1.327
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.08
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.84
    ],
    "hard-sol-b-judge_hard-05": [
     "w",
     1.027
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.809
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.986
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     1.129
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.883
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     1.124
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     44.397
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     37.508
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     39.405
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     37.912
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.396
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.6
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     1.297
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     1.203
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     1.279
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.086
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     1.123
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.95
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     1.045
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     1.687
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     1.729
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     1.442
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     1.372
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     1.405
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     1.456
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     1.526
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     1.227
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.227
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     32.441
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     31.596
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     32.459
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     2.194
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     32.859
    ]
   }
  },
  "jev-local": {
   "display": "jev-local (Qwen3.5-9B)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 81,
     "w": 15,
     "f": 0,
     "n": 0,
     "accuracy": 0.8438
    },
    "judge": {
     "c": 130,
     "w": 16,
     "f": 0,
     "n": 0,
     "accuracy": 0.8904
    },
    "hard": {
     "c": 130,
     "w": 90,
     "f": 0,
     "n": 0,
     "accuracy": 0.5909
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "intent": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "ordinal": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "routing": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     }
    },
    "judge": {
     "routing": {
      "c": 70,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8974
     },
     "adequacy": {
      "c": 60,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8824
     }
    },
    "hard": {
     "long_policy": {
      "c": 17,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.4474
     },
     "probability": {
      "c": 8,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "temporal_numeric": {
      "c": 8,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.2667
     },
     "ambiguous": {
      "c": 8,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5714
     },
     "multi_hop": {
      "c": 20,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5714
     },
     "tradeoff": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 23,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.697
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.196
    ],
    "easy-intent-01": [
     "c",
     0.49
    ],
    "easy-intent-02": [
     "c",
     1.188
    ],
    "easy-intent-03": [
     "c",
     1.194
    ],
    "easy-intent-04": [
     "c",
     1.185
    ],
    "easy-intent-05": [
     "c",
     1.183
    ],
    "easy-intent-06": [
     "c",
     2.722
    ],
    "easy-intent-07": [
     "c",
     2.552
    ],
    "easy-intent-08": [
     "c",
     1.851
    ],
    "easy-intent-09": [
     "c",
     1.168
    ],
    "easy-intent-10": [
     "c",
     1.166
    ],
    "easy-intent-11": [
     "c",
     0.483
    ],
    "easy-fact-00": [
     "c",
     1.039
    ],
    "easy-fact-01": [
     "c",
     1.025
    ],
    "easy-fact-02": [
     "c",
     1.038
    ],
    "easy-fact-03": [
     "c",
     1.033
    ],
    "easy-fact-04": [
     "c",
     0.341
    ],
    "easy-fact-05": [
     "c",
     1.028
    ],
    "easy-fact-06": [
     "c",
     1.044
    ],
    "easy-fact-07": [
     "c",
     1.038
    ],
    "easy-fact-08": [
     "c",
     1.033
    ],
    "easy-fact-09": [
     "c",
     0.342
    ],
    "easy-fact-10": [
     "c",
     0.341
    ],
    "easy-fact-11": [
     "c",
     0.341
    ],
    "easy-extraction-00": [
     "c",
     1.82
    ],
    "easy-extraction-01": [
     "c",
     1.13
    ],
    "easy-extraction-02": [
     "c",
     0.434
    ],
    "easy-extraction-03": [
     "c",
     0.435
    ],
    "easy-extraction-04": [
     "c",
     0.435
    ],
    "easy-extraction-05": [
     "c",
     0.434
    ],
    "easy-extraction-06": [
     "c",
     1.129
    ],
    "easy-extraction-07": [
     "c",
     1.123
    ],
    "easy-extraction-08": [
     "c",
     1.88
    ],
    "easy-extraction-09": [
     "c",
     1.885
    ],
    "easy-extraction-10": [
     "c",
     0.483
    ],
    "easy-extraction-11": [
     "c",
     0.436
    ],
    "easy-tool_selection-00": [
     "c",
     1.212
    ],
    "easy-tool_selection-01": [
     "c",
     1.185
    ],
    "easy-tool_selection-02": [
     "c",
     0.484
    ],
    "easy-tool_selection-03": [
     "c",
     1.187
    ],
    "easy-tool_selection-04": [
     "c",
     1.203
    ],
    "easy-tool_selection-05": [
     "c",
     0.483
    ],
    "easy-tool_selection-06": [
     "c",
     1.183
    ],
    "easy-tool_selection-07": [
     "c",
     0.483
    ],
    "easy-tool_selection-08": [
     "c",
     1.184
    ],
    "easy-tool_selection-09": [
     "c",
     1.183
    ],
    "easy-tool_selection-10": [
     "c",
     0.482
    ],
    "easy-tool_selection-11": [
     "c",
     0.482
    ],
    "original-policy-01-0": [
     "c",
     1.046
    ],
    "original-policy-01-1": [
     "c",
     1.047
    ],
    "original-policy-02-0": [
     "c",
     0.344
    ],
    "original-policy-02-1": [
     "w",
     1.041
    ],
    "original-policy-03-0": [
     "c",
     1.042
    ],
    "original-policy-03-1": [
     "c",
     1.072
    ],
    "original-policy-04-0": [
     "c",
     1.046
    ],
    "original-policy-04-1": [
     "c",
     1.044
    ],
    "original-policy-05-0": [
     "c",
     1.031
    ],
    "original-policy-05-1": [
     "c",
     1.025
    ],
    "original-policy-06-0": [
     "c",
     0.343
    ],
    "original-policy-06-1": [
     "c",
     1.028
    ],
    "original-intent-01-0": [
     "c",
     0.483
    ],
    "original-intent-01-1": [
     "c",
     0.482
    ],
    "original-intent-02-0": [
     "w",
     0.484
    ],
    "original-intent-02-1": [
     "c",
     0.483
    ],
    "original-intent-03-0": [
     "c",
     1.168
    ],
    "original-intent-03-1": [
     "c",
     0.484
    ],
    "original-intent-04-0": [
     "c",
     1.178
    ],
    "original-intent-04-1": [
     "c",
     0.483
    ],
    "original-intent-05-0": [
     "w",
     0.483
    ],
    "original-intent-05-1": [
     "w",
     1.173
    ],
    "original-intent-06-0": [
     "c",
     0.481
    ],
    "original-intent-06-1": [
     "c",
     0.483
    ],
    "original-ordinal-01-0": [
     "w",
     0.437
    ],
    "original-ordinal-01-1": [
     "c",
     0.435
    ],
    "original-ordinal-02-0": [
     "c",
     0.437
    ],
    "original-ordinal-02-1": [
     "c",
     0.438
    ],
    "original-ordinal-03-0": [
     "c",
     0.436
    ],
    "original-ordinal-03-1": [
     "c",
     0.435
    ],
    "original-ordinal-04-0": [
     "c",
     0.436
    ],
    "original-ordinal-04-1": [
     "c",
     0.435
    ],
    "original-ordinal-05-0": [
     "c",
     0.438
    ],
    "original-ordinal-05-1": [
     "c",
     0.436
    ],
    "original-ordinal-06-0": [
     "c",
     0.435
    ],
    "original-ordinal-06-1": [
     "c",
     0.435
    ],
    "original-extraction-01-0": [
     "c",
     1.113
    ],
    "original-extraction-01-1": [
     "c",
     0.436
    ],
    "original-extraction-02-0": [
     "c",
     0.436
    ],
    "original-extraction-02-1": [
     "c",
     0.437
    ],
    "original-extraction-03-0": [
     "c",
     0.436
    ],
    "original-extraction-03-1": [
     "c",
     0.437
    ],
    "original-extraction-04-0": [
     "c",
     0.435
    ],
    "original-extraction-04-1": [
     "c",
     0.435
    ],
    "original-extraction-05-0": [
     "c",
     0.435
    ],
    "original-extraction-05-1": [
     "c",
     0.435
    ],
    "original-extraction-06-0": [
     "c",
     0.437
    ],
    "original-extraction-06-1": [
     "c",
     0.436
    ],
    "original-adequacy-01-0": [
     "c",
     0.341
    ],
    "original-adequacy-01-1": [
     "c",
     0.342
    ],
    "original-adequacy-02-0": [
     "c",
     0.343
    ],
    "original-adequacy-02-1": [
     "c",
     0.344
    ],
    "original-adequacy-03-0": [
     "w",
     0.342
    ],
    "original-adequacy-03-1": [
     "w",
     0.343
    ],
    "original-adequacy-04-0": [
     "c",
     0.343
    ],
    "original-adequacy-04-1": [
     "c",
     0.344
    ],
    "original-adequacy-05-0": [
     "w",
     0.349
    ],
    "original-adequacy-05-1": [
     "w",
     0.349
    ],
    "original-adequacy-06-0": [
     "c",
     0.344
    ],
    "original-adequacy-06-1": [
     "c",
     0.342
    ],
    "original-routing-01-0": [
     "c",
     1.22
    ],
    "original-routing-01-1": [
     "c",
     0.53
    ],
    "original-routing-02-0": [
     "c",
     1.217
    ],
    "original-routing-02-1": [
     "c",
     0.532
    ],
    "original-routing-03-0": [
     "c",
     0.528
    ],
    "original-routing-03-1": [
     "c",
     1.906
    ],
    "original-routing-04-0": [
     "w",
     0.53
    ],
    "original-routing-04-1": [
     "w",
     0.531
    ],
    "original-routing-05-0": [
     "c",
     0.531
    ],
    "original-routing-05-1": [
     "w",
     0.53
    ],
    "original-routing-06-0": [
     "c",
     0.531
    ],
    "original-routing-06-1": [
     "c",
     0.53
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     5.134
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     3.628
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     3.892
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     3.269
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.403
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     1.415
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     3.462
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     1.396
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.058
    ],
    "hard-opus-a-probability-04": [
     "c",
     1.998
    ],
    "hard-opus-a-probability-07": [
     "w",
     2.014
    ],
    "hard-opus-a-probability-08": [
     "w",
     1.071
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     1.088
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     1.083
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     1.638
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     1.082
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     1.238
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     1.883
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     1.145
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.448
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     1.888
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     1.852
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     1.847
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     1.894
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     2.786
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     4.378
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     2.271
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     3.902
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     1.386
    ],
    "hard-opus-b-probability-01": [
     "w",
     2.176
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.453
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.494
    ],
    "hard-opus-b-probability-04": [
     "w",
     1.145
    ],
    "hard-opus-b-probability-06": [
     "w",
     1.245
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     2.133
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     1.287
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     1.124
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     2.745
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.394
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     2.708
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     4.901
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     3.91
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.728
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     2.386
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     2.388
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     3.818
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     1.737
    ],
    "hard-opus-c-probability-03": [
     "w",
     1.837
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     1.124
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     1.432
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.371
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     2.933
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     1.47
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     1.189
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     1.152
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.356
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     3.206
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.356
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.464
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.478
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     3.956
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     2.618
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     3.264
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     3.955
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     2.564
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     1.867
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     3.889
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     3.204
    ],
    "hard-sol-a-trap-02": [
     "c",
     1.053
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.351
    ],
    "hard-sol-a-trap-05": [
     "w",
     1.181
    ],
    "hard-sol-a-trap-06": [
     "c",
     1.863
    ],
    "hard-sol-a-trap-08": [
     "c",
     1.163
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.149
    ],
    "hard-sol-a-trap-13": [
     "c",
     1.045
    ],
    "hard-sol-a-trap-15": [
     "c",
     1.051
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.059
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.356
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.356
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     1.047
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     1.048
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     1.05
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.355
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     1.047
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     3.548
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     2.832
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     2.798
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     2.835
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.931
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.929
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     1.93
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.534
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.534
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.474
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     1.878
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     1.867
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.41
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.369
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     0.37
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.364
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.363
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     1.072
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.368
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     1.067
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     1.049
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.361
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     1.95
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     2.609
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     1.986
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     1.923
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.99
    ]
   }
  },
  "jqv": {
   "display": "jqv (Qwen3-32B zero-shot)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 92,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9583
    },
    "judge": {
     "c": 135,
     "w": 11,
     "f": 0,
     "n": 0,
     "accuracy": 0.9247
    },
    "hard": {
     "c": 142,
     "w": 78,
     "f": 0,
     "n": 0,
     "accuracy": 0.6455
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 72,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9231
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 19,
      "w": 19,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "probability": {
      "c": 12,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.6
     },
     "temporal_numeric": {
      "c": 7,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.2333
     },
     "ambiguous": {
      "c": 9,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6429
     },
     "multi_hop": {
      "c": 24,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.6857
     },
     "tradeoff": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 26,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.7879
     },
     "routing_hard": {
      "c": 9,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.768
    ],
    "easy-intent-01": [
     "c",
     0.299
    ],
    "easy-intent-02": [
     "c",
     0.729
    ],
    "easy-intent-03": [
     "c",
     0.732
    ],
    "easy-intent-04": [
     "c",
     0.73
    ],
    "easy-intent-05": [
     "c",
     0.732
    ],
    "easy-intent-06": [
     "c",
     0.733
    ],
    "easy-intent-07": [
     "c",
     0.733
    ],
    "easy-intent-08": [
     "c",
     0.742
    ],
    "easy-intent-09": [
     "c",
     0.727
    ],
    "easy-intent-10": [
     "c",
     0.728
    ],
    "easy-intent-11": [
     "c",
     0.295
    ],
    "easy-fact-00": [
     "c",
     0.752
    ],
    "easy-fact-01": [
     "c",
     0.72
    ],
    "easy-fact-02": [
     "c",
     0.729
    ],
    "easy-fact-03": [
     "c",
     0.725
    ],
    "easy-fact-04": [
     "c",
     0.293
    ],
    "easy-fact-05": [
     "c",
     0.74
    ],
    "easy-fact-06": [
     "c",
     0.296
    ],
    "easy-fact-07": [
     "c",
     0.743
    ],
    "easy-fact-08": [
     "c",
     0.724
    ],
    "easy-fact-09": [
     "c",
     0.292
    ],
    "easy-fact-10": [
     "c",
     0.317
    ],
    "easy-fact-11": [
     "c",
     0.75
    ],
    "easy-extraction-00": [
     "c",
     0.722
    ],
    "easy-extraction-01": [
     "c",
     0.725
    ],
    "easy-extraction-02": [
     "c",
     0.291
    ],
    "easy-extraction-03": [
     "c",
     0.74
    ],
    "easy-extraction-04": [
     "c",
     0.292
    ],
    "easy-extraction-05": [
     "c",
     0.317
    ],
    "easy-extraction-06": [
     "c",
     0.754
    ],
    "easy-extraction-07": [
     "c",
     0.717
    ],
    "easy-extraction-08": [
     "c",
     0.292
    ],
    "easy-extraction-09": [
     "c",
     0.308
    ],
    "easy-extraction-10": [
     "c",
     0.308
    ],
    "easy-extraction-11": [
     "c",
     0.74
    ],
    "easy-tool_selection-00": [
     "c",
     0.73
    ],
    "easy-tool_selection-01": [
     "c",
     0.297
    ],
    "easy-tool_selection-02": [
     "c",
     0.759
    ],
    "easy-tool_selection-03": [
     "c",
     0.731
    ],
    "easy-tool_selection-04": [
     "c",
     0.725
    ],
    "easy-tool_selection-05": [
     "c",
     0.298
    ],
    "easy-tool_selection-06": [
     "c",
     0.748
    ],
    "easy-tool_selection-07": [
     "c",
     0.296
    ],
    "easy-tool_selection-08": [
     "c",
     0.751
    ],
    "easy-tool_selection-09": [
     "c",
     0.299
    ],
    "easy-tool_selection-10": [
     "c",
     0.311
    ],
    "easy-tool_selection-11": [
     "c",
     0.31
    ],
    "original-policy-01-0": [
     "c",
     0.322
    ],
    "original-policy-01-1": [
     "c",
     0.319
    ],
    "original-policy-02-0": [
     "c",
     0.747
    ],
    "original-policy-02-1": [
     "c",
     0.296
    ],
    "original-policy-03-0": [
     "c",
     0.322
    ],
    "original-policy-03-1": [
     "w",
     0.755
    ],
    "original-policy-04-0": [
     "c",
     0.729
    ],
    "original-policy-04-1": [
     "c",
     0.732
    ],
    "original-policy-05-0": [
     "c",
     0.296
    ],
    "original-policy-05-1": [
     "c",
     0.749
    ],
    "original-policy-06-0": [
     "c",
     0.299
    ],
    "original-policy-06-1": [
     "c",
     0.311
    ],
    "original-intent-01-0": [
     "c",
     0.31
    ],
    "original-intent-01-1": [
     "c",
     0.317
    ],
    "original-intent-02-0": [
     "c",
     0.323
    ],
    "original-intent-02-1": [
     "c",
     0.317
    ],
    "original-intent-03-0": [
     "c",
     0.318
    ],
    "original-intent-03-1": [
     "c",
     0.311
    ],
    "original-intent-04-0": [
     "c",
     0.31
    ],
    "original-intent-04-1": [
     "c",
     0.31
    ],
    "original-intent-05-0": [
     "c",
     0.319
    ],
    "original-intent-05-1": [
     "c",
     0.745
    ],
    "original-intent-06-0": [
     "c",
     0.728
    ],
    "original-intent-06-1": [
     "c",
     0.295
    ],
    "original-ordinal-01-0": [
     "c",
     0.318
    ],
    "original-ordinal-01-1": [
     "c",
     0.32
    ],
    "original-ordinal-02-0": [
     "c",
     0.321
    ],
    "original-ordinal-02-1": [
     "c",
     0.324
    ],
    "original-ordinal-03-0": [
     "c",
     0.32
    ],
    "original-ordinal-03-1": [
     "c",
     0.297
    ],
    "original-ordinal-04-0": [
     "c",
     0.297
    ],
    "original-ordinal-04-1": [
     "c",
     0.298
    ],
    "original-ordinal-05-0": [
     "c",
     0.299
    ],
    "original-ordinal-05-1": [
     "c",
     0.299
    ],
    "original-ordinal-06-0": [
     "c",
     0.298
    ],
    "original-ordinal-06-1": [
     "c",
     0.734
    ],
    "original-extraction-01-0": [
     "c",
     0.728
    ],
    "original-extraction-01-1": [
     "c",
     0.295
    ],
    "original-extraction-02-0": [
     "c",
     0.318
    ],
    "original-extraction-02-1": [
     "c",
     0.308
    ],
    "original-extraction-03-0": [
     "c",
     0.738
    ],
    "original-extraction-03-1": [
     "c",
     0.294
    ],
    "original-extraction-04-0": [
     "c",
     0.309
    ],
    "original-extraction-04-1": [
     "c",
     0.308
    ],
    "original-extraction-05-0": [
     "c",
     0.309
    ],
    "original-extraction-05-1": [
     "c",
     0.31
    ],
    "original-extraction-06-0": [
     "c",
     0.318
    ],
    "original-extraction-06-1": [
     "c",
     0.318
    ],
    "original-adequacy-01-0": [
     "c",
     0.311
    ],
    "original-adequacy-01-1": [
     "c",
     0.316
    ],
    "original-adequacy-02-0": [
     "c",
     0.317
    ],
    "original-adequacy-02-1": [
     "c",
     0.317
    ],
    "original-adequacy-03-0": [
     "w",
     0.751
    ],
    "original-adequacy-03-1": [
     "c",
     0.318
    ],
    "original-adequacy-04-0": [
     "c",
     0.32
    ],
    "original-adequacy-04-1": [
     "c",
     0.312
    ],
    "original-adequacy-05-0": [
     "c",
     0.33
    ],
    "original-adequacy-05-1": [
     "w",
     0.323
    ],
    "original-adequacy-06-0": [
     "c",
     0.314
    ],
    "original-adequacy-06-1": [
     "c",
     0.319
    ],
    "original-routing-01-0": [
     "c",
     0.768
    ],
    "original-routing-01-1": [
     "c",
     0.73
    ],
    "original-routing-02-0": [
     "c",
     0.729
    ],
    "original-routing-02-1": [
     "c",
     0.759
    ],
    "original-routing-03-0": [
     "c",
     0.301
    ],
    "original-routing-03-1": [
     "c",
     0.766
    ],
    "original-routing-04-0": [
     "c",
     0.319
    ],
    "original-routing-04-1": [
     "c",
     0.32
    ],
    "original-routing-05-0": [
     "c",
     0.753
    ],
    "original-routing-05-1": [
     "c",
     0.322
    ],
    "original-routing-06-0": [
     "c",
     0.323
    ],
    "original-routing-06-1": [
     "c",
     0.315
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     1.69
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.225
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     1.189
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     1.206
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.178
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     1.185
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     1.219
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     1.177
    ],
    "hard-opus-a-probability-03": [
     "w",
     0.786
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.903
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.505
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.782
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.794
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.805
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.951
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.799
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.808
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.799
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.793
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.792
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.814
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.369
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.808
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.811
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     1.316
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     1.151
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     1.214
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     1.216
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     1.177
    ],
    "hard-opus-b-probability-01": [
     "w",
     1.059
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.796
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.923
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.808
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.885
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     0.853
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.837
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.842
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.826
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.811
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.808
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     1.574
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     1.55
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.522
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.524
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.49
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     1.608
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     1.568
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.803
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.869
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.853
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.84
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.879
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.881
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.844
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.753
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.751
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.773
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.772
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.316
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.803
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.8
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     1.231
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     1.216
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     1.206
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     1.213
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.785
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     1.225
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     1.212
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.34
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.743
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.768
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.77
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.755
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.747
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.315
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.792
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.345
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.767
    ],
    "hard-sol-b-judge_hard-05": [
     "w",
     0.318
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.775
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.316
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.339
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.33
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     0.774
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     1.408
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     1.361
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.381
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     1.383
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.775
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.791
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.771
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.769
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.772
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.339
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.343
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.318
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.857
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.354
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.351
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     0.34
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.783
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.341
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.796
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.772
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     0.769
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.34
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     1.242
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     1.214
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     1.22
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.803
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.218
    ]
   }
  },
  "kev-0.5b": {
   "display": "kev 0.5B",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 69,
     "w": 3,
     "f": 0,
     "n": 0,
     "accuracy": 0.9583
    },
    "standard": {
     "c": 50,
     "w": 46,
     "f": 0,
     "n": 0,
     "accuracy": 0.5208
    },
    "judge": {
     "c": 104,
     "w": 42,
     "f": 0,
     "n": 0,
     "accuracy": 0.7123
    },
    "hard": {
     "c": 68,
     "w": 152,
     "f": 0,
     "n": 0,
     "accuracy": 0.3091
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 17,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9444
     },
     "fact": {
      "c": 16,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8889
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "intent": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "ordinal": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "extraction": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adequacy": {
      "c": 5,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.3125
     },
     "routing": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     }
    },
    "judge": {
     "routing": {
      "c": 42,
      "w": 36,
      "f": 0,
      "n": 0,
      "accuracy": 0.5385
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 7,
      "w": 31,
      "f": 0,
      "n": 0,
      "accuracy": 0.1842
     },
     "probability": {
      "c": 4,
      "w": 16,
      "f": 0,
      "n": 0,
      "accuracy": 0.2
     },
     "temporal_numeric": {
      "c": 4,
      "w": 26,
      "f": 0,
      "n": 0,
      "accuracy": 0.1333
     },
     "ambiguous": {
      "c": 4,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.2857
     },
     "multi_hop": {
      "c": 11,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.3143
     },
     "tradeoff": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "adversarial": {
      "c": 4,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "trap": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "judge_hard": {
      "c": 20,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.6061
     },
     "routing_hard": {
      "c": 0,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.81
    ],
    "easy-intent-01": [
     "c",
     0.46
    ],
    "easy-intent-02": [
     "c",
     0.588
    ],
    "easy-intent-03": [
     "c",
     0.596
    ],
    "easy-intent-04": [
     "c",
     0.593
    ],
    "easy-intent-05": [
     "c",
     0.756
    ],
    "easy-intent-06": [
     "c",
     0.547
    ],
    "easy-intent-07": [
     "c",
     0.433
    ],
    "easy-intent-08": [
     "c",
     0.561
    ],
    "easy-intent-09": [
     "w",
     0.59
    ],
    "easy-intent-10": [
     "c",
     0.599
    ],
    "easy-intent-11": [
     "c",
     0.924
    ],
    "easy-fact-00": [
     "c",
     0.587
    ],
    "easy-fact-01": [
     "c",
     0.593
    ],
    "easy-fact-02": [
     "c",
     0.606
    ],
    "easy-fact-03": [
     "c",
     0.587
    ],
    "easy-fact-04": [
     "c",
     0.615
    ],
    "easy-fact-05": [
     "w",
     0.777
    ],
    "easy-fact-06": [
     "c",
     0.536
    ],
    "easy-fact-07": [
     "c",
     0.395
    ],
    "easy-fact-08": [
     "c",
     0.382
    ],
    "easy-fact-09": [
     "c",
     1.054
    ],
    "easy-fact-10": [
     "c",
     1.307
    ],
    "easy-fact-11": [
     "c",
     0.43
    ],
    "easy-extraction-00": [
     "c",
     0.561
    ],
    "easy-extraction-01": [
     "c",
     0.428
    ],
    "easy-extraction-02": [
     "c",
     0.557
    ],
    "easy-extraction-03": [
     "c",
     0.422
    ],
    "easy-extraction-04": [
     "c",
     0.526
    ],
    "easy-extraction-05": [
     "c",
     0.902
    ],
    "easy-extraction-06": [
     "c",
     0.543
    ],
    "easy-extraction-07": [
     "c",
     0.603
    ],
    "easy-extraction-08": [
     "c",
     0.599
    ],
    "easy-extraction-09": [
     "c",
     0.428
    ],
    "easy-extraction-10": [
     "c",
     0.559
    ],
    "easy-extraction-11": [
     "c",
     0.419
    ],
    "easy-tool_selection-00": [
     "c",
     0.379
    ],
    "easy-tool_selection-01": [
     "c",
     0.89
    ],
    "easy-tool_selection-02": [
     "c",
     0.435
    ],
    "easy-tool_selection-03": [
     "c",
     0.456
    ],
    "easy-tool_selection-04": [
     "c",
     0.389
    ],
    "easy-tool_selection-05": [
     "c",
     0.541
    ],
    "easy-tool_selection-06": [
     "c",
     0.385
    ],
    "easy-tool_selection-07": [
     "c",
     0.456
    ],
    "easy-tool_selection-08": [
     "c",
     0.349
    ],
    "easy-tool_selection-09": [
     "c",
     0.377
    ],
    "easy-tool_selection-10": [
     "c",
     0.381
    ],
    "easy-tool_selection-11": [
     "c",
     0.389
    ],
    "original-policy-01-0": [
     "c",
     0.437
    ],
    "original-policy-01-1": [
     "c",
     0.379
    ],
    "original-policy-02-0": [
     "w",
     0.551
    ],
    "original-policy-02-1": [
     "w",
     0.433
    ],
    "original-policy-03-0": [
     "w",
     0.374
    ],
    "original-policy-03-1": [
     "w",
     0.388
    ],
    "original-policy-04-0": [
     "c",
     0.386
    ],
    "original-policy-04-1": [
     "c",
     0.387
    ],
    "original-policy-05-0": [
     "c",
     0.375
    ],
    "original-policy-05-1": [
     "c",
     0.288
    ],
    "original-policy-06-0": [
     "w",
     1.103
    ],
    "original-policy-06-1": [
     "c",
     0.383
    ],
    "original-intent-01-0": [
     "w",
     0.742
    ],
    "original-intent-01-1": [
     "w",
     0.39
    ],
    "original-intent-02-0": [
     "w",
     0.388
    ],
    "original-intent-02-1": [
     "w",
     0.387
    ],
    "original-intent-03-0": [
     "w",
     0.542
    ],
    "original-intent-03-1": [
     "c",
     0.368
    ],
    "original-intent-04-0": [
     "c",
     0.371
    ],
    "original-intent-04-1": [
     "c",
     0.372
    ],
    "original-intent-05-0": [
     "c",
     0.359
    ],
    "original-intent-05-1": [
     "c",
     0.553
    ],
    "original-intent-06-0": [
     "w",
     0.43
    ],
    "original-intent-06-1": [
     "w",
     0.38
    ],
    "original-ordinal-01-0": [
     "c",
     0.38
    ],
    "original-ordinal-01-1": [
     "c",
     0.395
    ],
    "original-ordinal-02-0": [
     "c",
     0.386
    ],
    "original-ordinal-02-1": [
     "c",
     0.388
    ],
    "original-ordinal-03-0": [
     "c",
     0.381
    ],
    "original-ordinal-03-1": [
     "c",
     0.376
    ],
    "original-ordinal-04-0": [
     "w",
     0.388
    ],
    "original-ordinal-04-1": [
     "w",
     0.371
    ],
    "original-ordinal-05-0": [
     "w",
     0.387
    ],
    "original-ordinal-05-1": [
     "w",
     0.387
    ],
    "original-ordinal-06-0": [
     "c",
     0.385
    ],
    "original-ordinal-06-1": [
     "c",
     0.729
    ],
    "original-extraction-01-0": [
     "w",
     0.375
    ],
    "original-extraction-01-1": [
     "w",
     0.379
    ],
    "original-extraction-02-0": [
     "c",
     0.385
    ],
    "original-extraction-02-1": [
     "c",
     0.384
    ],
    "original-extraction-03-0": [
     "c",
     0.454
    ],
    "original-extraction-03-1": [
     "c",
     0.742
    ],
    "original-extraction-04-0": [
     "c",
     0.374
    ],
    "original-extraction-04-1": [
     "w",
     0.946
    ],
    "original-extraction-05-0": [
     "c",
     0.373
    ],
    "original-extraction-05-1": [
     "c",
     0.559
    ],
    "original-extraction-06-0": [
     "w",
     0.588
    ],
    "original-extraction-06-1": [
     "w",
     0.597
    ],
    "original-adequacy-01-0": [
     "w",
     0.596
    ],
    "original-adequacy-01-1": [
     "w",
     0.597
    ],
    "original-adequacy-02-0": [
     "c",
     0.567
    ],
    "original-adequacy-02-1": [
     "c",
     0.606
    ],
    "original-adequacy-03-0": [
     "w",
     0.61
    ],
    "original-adequacy-03-1": [
     "w",
     0.583
    ],
    "original-adequacy-04-0": [
     "w",
     0.603
    ],
    "original-adequacy-04-1": [
     "w",
     0.597
    ],
    "original-adequacy-05-0": [
     "c",
     0.585
    ],
    "original-adequacy-05-1": [
     "w",
     0.602
    ],
    "original-adequacy-06-0": [
     "w",
     0.593
    ],
    "original-adequacy-06-1": [
     "w",
     0.584
    ],
    "original-routing-01-0": [
     "c",
     0.922
    ],
    "original-routing-01-1": [
     "w",
     0.6
    ],
    "original-routing-02-0": [
     "w",
     0.592
    ],
    "original-routing-02-1": [
     "w",
     0.604
    ],
    "original-routing-03-0": [
     "c",
     0.6
    ],
    "original-routing-03-1": [
     "c",
     0.596
    ],
    "original-routing-04-0": [
     "w",
     0.586
    ],
    "original-routing-04-1": [
     "w",
     0.921
    ],
    "original-routing-05-0": [
     "w",
     0.592
    ],
    "original-routing-05-1": [
     "w",
     0.595
    ],
    "original-routing-06-0": [
     "c",
     0.596
    ],
    "original-routing-06-1": [
     "c",
     0.432
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     0.942
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.068
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     0.753
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.885
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.897
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     0.784
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     1.004
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.645
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.38
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.889
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.72
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.462
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.589
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.597
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.605
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.6
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.78
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.458
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.754
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.79
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.587
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.604
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     1.376
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.555
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.779
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.754
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.755
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     1.404
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.575
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.597
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.378
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.544
    ],
    "hard-opus-b-probability-04": [
     "w",
     0.426
    ],
    "hard-opus-b-probability-06": [
     "w",
     1.362
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.417
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.552
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.748
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.591
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.462
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     0.6
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     1.056
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     0.909
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.834
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.17
    ],
    "hard-opus-c-long_policy-08": [
     "c",
     0.916
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     1.442
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.999
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.59
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.766
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.372
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.079
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.56
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.545
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.68
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.35
    ],
    "hard-sol-a-adversarial-06": [
     "w",
     0.554
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.587
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.407
    ],
    "hard-sol-a-adversarial-09": [
     "w",
     0.383
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.372
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.842
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.71
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     1.221
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.814
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.688
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     0.754
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.576
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.564
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.584
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.376
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.546
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.557
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.427
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.549
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.585
    ],
    "hard-sol-a-trap-15": [
     "w",
     0.599
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.425
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.726
    ],
    "hard-sol-b-judge_hard-05": [
     "w",
     0.374
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.379
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.386
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.562
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.592
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.427
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     1.362
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     1.085
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.71
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     1.309
    ],
    "hard-sol-b-routing_hard-01": [
     "w",
     0.525
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     0.379
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     0.528
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.545
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     0.364
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.368
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.375
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.546
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.601
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.745
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.897
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.364
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.589
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.466
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.368
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.548
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.587
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.424
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     0.681
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     0.832
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     0.851
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.598
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.406
    ]
   }
  },
  "kev-0.6b": {
   "display": "kev 0.6B (research preview)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 78,
     "w": 18,
     "f": 0,
     "n": 0,
     "accuracy": 0.8125
    },
    "judge": {
     "c": 97,
     "w": 49,
     "f": 0,
     "n": 0,
     "accuracy": 0.6644
    },
    "hard": {
     "c": 88,
     "w": 132,
     "f": 0,
     "n": 0,
     "accuracy": 0.4
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "intent": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "ordinal": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "extraction": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "adequacy": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 35,
      "w": 43,
      "f": 0,
      "n": 0,
      "accuracy": 0.4487
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 11,
      "w": 27,
      "f": 0,
      "n": 0,
      "accuracy": 0.2895
     },
     "probability": {
      "c": 6,
      "w": 14,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "temporal_numeric": {
      "c": 10,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "ambiguous": {
      "c": 4,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.2857
     },
     "multi_hop": {
      "c": 11,
      "w": 24,
      "f": 0,
      "n": 0,
      "accuracy": 0.3143
     },
     "tradeoff": {
      "c": 6,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 5,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.646
    ],
    "easy-intent-01": [
     "c",
     0.56
    ],
    "easy-intent-02": [
     "c",
     0.938
    ],
    "easy-intent-03": [
     "c",
     0.97
    ],
    "easy-intent-04": [
     "c",
     0.587
    ],
    "easy-intent-05": [
     "c",
     0.634
    ],
    "easy-intent-06": [
     "c",
     0.416
    ],
    "easy-intent-07": [
     "c",
     0.373
    ],
    "easy-intent-08": [
     "c",
     0.383
    ],
    "easy-intent-09": [
     "c",
     0.384
    ],
    "easy-intent-10": [
     "c",
     0.373
    ],
    "easy-intent-11": [
     "c",
     0.375
    ],
    "easy-fact-00": [
     "c",
     0.371
    ],
    "easy-fact-01": [
     "c",
     0.372
    ],
    "easy-fact-02": [
     "c",
     0.313
    ],
    "easy-fact-03": [
     "c",
     0.374
    ],
    "easy-fact-04": [
     "c",
     0.359
    ],
    "easy-fact-05": [
     "c",
     0.639
    ],
    "easy-fact-06": [
     "c",
     0.575
    ],
    "easy-fact-07": [
     "c",
     0.904
    ],
    "easy-fact-08": [
     "c",
     0.404
    ],
    "easy-fact-09": [
     "c",
     0.561
    ],
    "easy-fact-10": [
     "c",
     0.417
    ],
    "easy-fact-11": [
     "c",
     0.868
    ],
    "easy-extraction-00": [
     "c",
     0.635
    ],
    "easy-extraction-01": [
     "c",
     0.952
    ],
    "easy-extraction-02": [
     "c",
     0.426
    ],
    "easy-extraction-03": [
     "c",
     0.506
    ],
    "easy-extraction-04": [
     "c",
     0.419
    ],
    "easy-extraction-05": [
     "c",
     0.537
    ],
    "easy-extraction-06": [
     "c",
     0.596
    ],
    "easy-extraction-07": [
     "c",
     0.419
    ],
    "easy-extraction-08": [
     "c",
     0.76
    ],
    "easy-extraction-09": [
     "c",
     0.384
    ],
    "easy-extraction-10": [
     "c",
     0.755
    ],
    "easy-extraction-11": [
     "c",
     0.357
    ],
    "easy-tool_selection-00": [
     "c",
     0.313
    ],
    "easy-tool_selection-01": [
     "c",
     0.551
    ],
    "easy-tool_selection-02": [
     "c",
     0.418
    ],
    "easy-tool_selection-03": [
     "c",
     0.378
    ],
    "easy-tool_selection-04": [
     "c",
     0.549
    ],
    "easy-tool_selection-05": [
     "c",
     0.424
    ],
    "easy-tool_selection-06": [
     "c",
     0.759
    ],
    "easy-tool_selection-07": [
     "c",
     0.386
    ],
    "easy-tool_selection-08": [
     "c",
     0.461
    ],
    "easy-tool_selection-09": [
     "c",
     0.365
    ],
    "easy-tool_selection-10": [
     "c",
     0.382
    ],
    "easy-tool_selection-11": [
     "c",
     0.391
    ],
    "original-policy-01-0": [
     "c",
     0.41
    ],
    "original-policy-01-1": [
     "c",
     0.366
    ],
    "original-policy-02-0": [
     "c",
     0.382
    ],
    "original-policy-02-1": [
     "w",
     0.379
    ],
    "original-policy-03-0": [
     "c",
     0.383
    ],
    "original-policy-03-1": [
     "c",
     0.365
    ],
    "original-policy-04-0": [
     "c",
     0.382
    ],
    "original-policy-04-1": [
     "c",
     0.387
    ],
    "original-policy-05-0": [
     "c",
     0.377
    ],
    "original-policy-05-1": [
     "c",
     0.383
    ],
    "original-policy-06-0": [
     "c",
     0.37
    ],
    "original-policy-06-1": [
     "c",
     0.373
    ],
    "original-intent-01-0": [
     "w",
     0.547
    ],
    "original-intent-01-1": [
     "w",
     0.432
    ],
    "original-intent-02-0": [
     "w",
     0.97
    ],
    "original-intent-02-1": [
     "c",
     0.377
    ],
    "original-intent-03-0": [
     "w",
     0.466
    ],
    "original-intent-03-1": [
     "c",
     0.379
    ],
    "original-intent-04-0": [
     "c",
     0.373
    ],
    "original-intent-04-1": [
     "c",
     0.559
    ],
    "original-intent-05-0": [
     "c",
     0.379
    ],
    "original-intent-05-1": [
     "c",
     0.708
    ],
    "original-intent-06-0": [
     "c",
     0.386
    ],
    "original-intent-06-1": [
     "c",
     0.378
    ],
    "original-ordinal-01-0": [
     "c",
     0.547
    ],
    "original-ordinal-01-1": [
     "c",
     0.605
    ],
    "original-ordinal-02-0": [
     "c",
     0.438
    ],
    "original-ordinal-02-1": [
     "c",
     0.385
    ],
    "original-ordinal-03-0": [
     "c",
     0.379
    ],
    "original-ordinal-03-1": [
     "c",
     3.194
    ],
    "original-ordinal-04-0": [
     "c",
     0.363
    ],
    "original-ordinal-04-1": [
     "c",
     0.383
    ],
    "original-ordinal-05-0": [
     "w",
     0.717
    ],
    "original-ordinal-05-1": [
     "w",
     0.418
    ],
    "original-ordinal-06-0": [
     "c",
     0.538
    ],
    "original-ordinal-06-1": [
     "c",
     0.606
    ],
    "original-extraction-01-0": [
     "w",
     0.602
    ],
    "original-extraction-01-1": [
     "w",
     0.607
    ],
    "original-extraction-02-0": [
     "c",
     0.58
    ],
    "original-extraction-02-1": [
     "c",
     0.607
    ],
    "original-extraction-03-0": [
     "c",
     0.605
    ],
    "original-extraction-03-1": [
     "c",
     0.596
    ],
    "original-extraction-04-0": [
     "c",
     0.593
    ],
    "original-extraction-04-1": [
     "c",
     0.597
    ],
    "original-extraction-05-0": [
     "c",
     0.59
    ],
    "original-extraction-05-1": [
     "c",
     0.599
    ],
    "original-extraction-06-0": [
     "c",
     0.749
    ],
    "original-extraction-06-1": [
     "c",
     0.593
    ],
    "original-adequacy-01-0": [
     "c",
     0.607
    ],
    "original-adequacy-01-1": [
     "c",
     0.595
    ],
    "original-adequacy-02-0": [
     "w",
     0.593
    ],
    "original-adequacy-02-1": [
     "c",
     0.606
    ],
    "original-adequacy-03-0": [
     "w",
     0.6
    ],
    "original-adequacy-03-1": [
     "w",
     0.787
    ],
    "original-adequacy-04-0": [
     "c",
     0.588
    ],
    "original-adequacy-04-1": [
     "c",
     0.594
    ],
    "original-adequacy-05-0": [
     "w",
     0.594
    ],
    "original-adequacy-05-1": [
     "w",
     0.596
    ],
    "original-adequacy-06-0": [
     "c",
     0.595
    ],
    "original-adequacy-06-1": [
     "c",
     0.594
    ],
    "original-routing-01-0": [
     "c",
     0.591
    ],
    "original-routing-01-1": [
     "c",
     0.962
    ],
    "original-routing-02-0": [
     "c",
     0.604
    ],
    "original-routing-02-1": [
     "c",
     0.571
    ],
    "original-routing-03-0": [
     "c",
     0.984
    ],
    "original-routing-03-1": [
     "c",
     0.95
    ],
    "original-routing-04-0": [
     "c",
     0.585
    ],
    "original-routing-04-1": [
     "c",
     0.587
    ],
    "original-routing-05-0": [
     "c",
     1.531
    ],
    "original-routing-05-1": [
     "c",
     0.567
    ],
    "original-routing-06-0": [
     "c",
     0.599
    ],
    "original-routing-06-1": [
     "c",
     0.596
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     1.467
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.729
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     1.044
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     1.277
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.89
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     0.696
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.738
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.75
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.911
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.6
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.718
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.59
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.699
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.581
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     0.794
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.571
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     0.431
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.541
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.757
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.888
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.687
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.531
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.784
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.405
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.873
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.996
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.936
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.803
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     0.824
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.769
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.423
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.395
    ],
    "hard-opus-b-probability-04": [
     "w",
     0.368
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.37
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.539
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.608
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.598
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.608
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     0.519
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.421
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     0.798
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     0.821
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.153
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.943
    ],
    "hard-opus-c-long_policy-08": [
     "c",
     0.769
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     0.838
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.949
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.351
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.549
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.552
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.391
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.381
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.877
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.909
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.575
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.582
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.587
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.59
    ],
    "hard-sol-a-adversarial-09": [
     "w",
     0.902
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.589
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.74
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.584
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     0.887
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     1.93
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.875
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     0.769
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     0.767
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     1.308
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.735
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.365
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.378
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.383
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.369
    ],
    "hard-sol-a-trap-10": [
     "w",
     0.364
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.377
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.371
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.369
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.34
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.384
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.367
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.379
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.53
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.568
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.382
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     1.034
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     1.38
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.998
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.931
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.377
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.035
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     0.568
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.575
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.579
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.417
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.554
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.597
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     1.316
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.909
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     1.087
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.58
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.936
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.549
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.589
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.742
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.589
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.599
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     1.176
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.922
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     0.999
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.36
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.49
    ]
   }
  },
  "kev-4b": {
   "display": "kev 4B (research preview)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 88,
     "w": 8,
     "f": 0,
     "n": 0,
     "accuracy": 0.9167
    },
    "judge": {
     "c": 125,
     "w": 21,
     "f": 0,
     "n": 0,
     "accuracy": 0.8562
    },
    "hard": {
     "c": 93,
     "w": 127,
     "f": 0,
     "n": 0,
     "accuracy": 0.4227
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "adequacy": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 63,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.8077
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 9,
      "w": 29,
      "f": 0,
      "n": 0,
      "accuracy": 0.2368
     },
     "probability": {
      "c": 9,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.45
     },
     "temporal_numeric": {
      "c": 3,
      "w": 27,
      "f": 0,
      "n": 0,
      "accuracy": 0.1
     },
     "ambiguous": {
      "c": 2,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.1429
     },
     "multi_hop": {
      "c": 16,
      "w": 19,
      "f": 0,
      "n": 0,
      "accuracy": 0.4571
     },
     "tradeoff": {
      "c": 3,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "adversarial": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "trap": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.829
    ],
    "easy-intent-01": [
     "c",
     0.753
    ],
    "easy-intent-02": [
     "c",
     0.593
    ],
    "easy-intent-03": [
     "c",
     0.588
    ],
    "easy-intent-04": [
     "c",
     0.933
    ],
    "easy-intent-05": [
     "c",
     0.477
    ],
    "easy-intent-06": [
     "c",
     0.435
    ],
    "easy-intent-07": [
     "c",
     0.385
    ],
    "easy-intent-08": [
     "c",
     0.388
    ],
    "easy-intent-09": [
     "c",
     0.375
    ],
    "easy-intent-10": [
     "c",
     0.391
    ],
    "easy-intent-11": [
     "c",
     0.39
    ],
    "easy-fact-00": [
     "c",
     0.384
    ],
    "easy-fact-01": [
     "c",
     0.715
    ],
    "easy-fact-02": [
     "c",
     0.482
    ],
    "easy-fact-03": [
     "c",
     0.611
    ],
    "easy-fact-04": [
     "c",
     0.424
    ],
    "easy-fact-05": [
     "c",
     0.735
    ],
    "easy-fact-06": [
     "c",
     0.406
    ],
    "easy-fact-07": [
     "c",
     0.377
    ],
    "easy-fact-08": [
     "c",
     0.476
    ],
    "easy-fact-09": [
     "c",
     0.391
    ],
    "easy-fact-10": [
     "c",
     0.388
    ],
    "easy-fact-11": [
     "c",
     0.549
    ],
    "easy-extraction-00": [
     "c",
     0.401
    ],
    "easy-extraction-01": [
     "c",
     0.534
    ],
    "easy-extraction-02": [
     "c",
     0.661
    ],
    "easy-extraction-03": [
     "c",
     0.764
    ],
    "easy-extraction-04": [
     "c",
     0.372
    ],
    "easy-extraction-05": [
     "c",
     0.865
    ],
    "easy-extraction-06": [
     "c",
     0.413
    ],
    "easy-extraction-07": [
     "c",
     0.375
    ],
    "easy-extraction-08": [
     "c",
     0.389
    ],
    "easy-extraction-09": [
     "c",
     0.383
    ],
    "easy-extraction-10": [
     "c",
     0.384
    ],
    "easy-extraction-11": [
     "c",
     0.762
    ],
    "easy-tool_selection-00": [
     "c",
     0.438
    ],
    "easy-tool_selection-01": [
     "c",
     0.369
    ],
    "easy-tool_selection-02": [
     "c",
     0.364
    ],
    "easy-tool_selection-03": [
     "c",
     0.381
    ],
    "easy-tool_selection-04": [
     "c",
     0.542
    ],
    "easy-tool_selection-05": [
     "c",
     0.646
    ],
    "easy-tool_selection-06": [
     "c",
     0.426
    ],
    "easy-tool_selection-07": [
     "c",
     0.378
    ],
    "easy-tool_selection-08": [
     "c",
     0.39
    ],
    "easy-tool_selection-09": [
     "c",
     0.345
    ],
    "easy-tool_selection-10": [
     "c",
     0.997
    ],
    "easy-tool_selection-11": [
     "c",
     0.731
    ],
    "original-policy-01-0": [
     "c",
     0.652
    ],
    "original-policy-01-1": [
     "c",
     0.48
    ],
    "original-policy-02-0": [
     "c",
     0.369
    ],
    "original-policy-02-1": [
     "c",
     0.531
    ],
    "original-policy-03-0": [
     "c",
     0.614
    ],
    "original-policy-03-1": [
     "c",
     0.799
    ],
    "original-policy-04-0": [
     "c",
     0.382
    ],
    "original-policy-04-1": [
     "c",
     0.384
    ],
    "original-policy-05-0": [
     "c",
     0.388
    ],
    "original-policy-05-1": [
     "c",
     0.375
    ],
    "original-policy-06-0": [
     "c",
     0.383
    ],
    "original-policy-06-1": [
     "c",
     0.754
    ],
    "original-intent-01-0": [
     "c",
     0.565
    ],
    "original-intent-01-1": [
     "w",
     0.438
    ],
    "original-intent-02-0": [
     "c",
     0.392
    ],
    "original-intent-02-1": [
     "c",
     0.382
    ],
    "original-intent-03-0": [
     "c",
     0.396
    ],
    "original-intent-03-1": [
     "c",
     0.73
    ],
    "original-intent-04-0": [
     "c",
     0.953
    ],
    "original-intent-04-1": [
     "w",
     0.526
    ],
    "original-intent-05-0": [
     "c",
     0.433
    ],
    "original-intent-05-1": [
     "c",
     0.388
    ],
    "original-intent-06-0": [
     "c",
     0.718
    ],
    "original-intent-06-1": [
     "c",
     0.381
    ],
    "original-ordinal-01-0": [
     "c",
     0.377
    ],
    "original-ordinal-01-1": [
     "c",
     0.93
    ],
    "original-ordinal-02-0": [
     "c",
     1.168
    ],
    "original-ordinal-02-1": [
     "c",
     0.651
    ],
    "original-ordinal-03-0": [
     "c",
     0.946
    ],
    "original-ordinal-03-1": [
     "c",
     0.828
    ],
    "original-ordinal-04-0": [
     "c",
     0.61
    ],
    "original-ordinal-04-1": [
     "c",
     0.56
    ],
    "original-ordinal-05-0": [
     "c",
     0.599
    ],
    "original-ordinal-05-1": [
     "c",
     0.596
    ],
    "original-ordinal-06-0": [
     "c",
     0.605
    ],
    "original-ordinal-06-1": [
     "c",
     0.42
    ],
    "original-extraction-01-0": [
     "c",
     0.385
    ],
    "original-extraction-01-1": [
     "w",
     0.377
    ],
    "original-extraction-02-0": [
     "c",
     0.542
    ],
    "original-extraction-02-1": [
     "c",
     0.421
    ],
    "original-extraction-03-0": [
     "c",
     0.538
    ],
    "original-extraction-03-1": [
     "c",
     0.383
    ],
    "original-extraction-04-0": [
     "c",
     0.555
    ],
    "original-extraction-04-1": [
     "c",
     0.442
    ],
    "original-extraction-05-0": [
     "c",
     0.382
    ],
    "original-extraction-05-1": [
     "c",
     0.715
    ],
    "original-extraction-06-0": [
     "c",
     0.597
    ],
    "original-extraction-06-1": [
     "c",
     0.767
    ],
    "original-adequacy-01-0": [
     "c",
     0.311
    ],
    "original-adequacy-01-1": [
     "c",
     0.855
    ],
    "original-adequacy-02-0": [
     "w",
     0.354
    ],
    "original-adequacy-02-1": [
     "c",
     0.53
    ],
    "original-adequacy-03-0": [
     "w",
     0.604
    ],
    "original-adequacy-03-1": [
     "w",
     0.43
    ],
    "original-adequacy-04-0": [
     "c",
     0.546
    ],
    "original-adequacy-04-1": [
     "c",
     0.99
    ],
    "original-adequacy-05-0": [
     "w",
     0.553
    ],
    "original-adequacy-05-1": [
     "w",
     0.943
    ],
    "original-adequacy-06-0": [
     "c",
     0.415
    ],
    "original-adequacy-06-1": [
     "c",
     0.369
    ],
    "original-routing-01-0": [
     "c",
     0.377
    ],
    "original-routing-01-1": [
     "c",
     0.373
    ],
    "original-routing-02-0": [
     "c",
     0.384
    ],
    "original-routing-02-1": [
     "c",
     0.373
    ],
    "original-routing-03-0": [
     "c",
     0.564
    ],
    "original-routing-03-1": [
     "c",
     0.605
    ],
    "original-routing-04-0": [
     "c",
     0.593
    ],
    "original-routing-04-1": [
     "c",
     0.612
    ],
    "original-routing-05-0": [
     "c",
     0.781
    ],
    "original-routing-05-1": [
     "c",
     0.586
    ],
    "original-routing-06-0": [
     "c",
     0.683
    ],
    "original-routing-06-1": [
     "c",
     0.537
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     1.935
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.118
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     1.184
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     1.189
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.389
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     1.096
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     1.7
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     1.472
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.388
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.534
    ],
    "hard-opus-a-probability-07": [
     "w",
     1.044
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.392
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.864
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.405
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.575
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.423
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.591
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.461
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.598
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.894
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     1.501
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.511
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     0.405
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.437
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     1.207
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     1.007
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.97
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.948
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     1.176
    ],
    "hard-opus-b-probability-01": [
     "c",
     1.038
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.42
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.518
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.603
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.708
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.637
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.465
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.41
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.637
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.494
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     0.945
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     2.25
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     1.775
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.629
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.715
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.466
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     1.587
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     1.887
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.656
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     1.146
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     3.272
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.155
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     1.182
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.822
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.699
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.563
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.576
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.599
    ],
    "hard-sol-a-adversarial-08": [
     "w",
     0.571
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.597
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.767
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     1.208
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     1.031
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     1.598
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     1.018
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     1.217
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     1.7
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     2.004
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     1.146
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.78
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.435
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.387
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.576
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.42
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.363
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.889
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.498
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.61
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.573
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.586
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.418
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.369
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.866
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.593
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.484
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     1.458
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     1.729
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.908
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     1.498
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.325
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.363
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.366
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.387
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.607
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.744
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     0.765
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.558
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.448
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.499
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.408
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.916
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.377
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.729
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.39
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     1.01
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.602
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.37
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     1.105
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     1.645
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     1.271
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.933
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.334
    ]
   }
  },
  "kev-8b": {
   "display": "kev 8B (research preview)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 89,
     "w": 7,
     "f": 0,
     "n": 0,
     "accuracy": 0.9271
    },
    "judge": {
     "c": 132,
     "w": 14,
     "f": 0,
     "n": 0,
     "accuracy": 0.9041
    },
    "hard": {
     "c": 104,
     "w": 116,
     "f": 0,
     "n": 0,
     "accuracy": 0.4727
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "adequacy": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 70,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8974
     },
     "adequacy": {
      "c": 62,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9118
     }
    },
    "hard": {
     "long_policy": {
      "c": 15,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.3947
     },
     "probability": {
      "c": 10,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "temporal_numeric": {
      "c": 0,
      "w": 30,
      "f": 0,
      "n": 0,
      "accuracy": 0.0
     },
     "ambiguous": {
      "c": 3,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.2143
     },
     "multi_hop": {
      "c": 18,
      "w": 17,
      "f": 0,
      "n": 0,
      "accuracy": 0.5143
     },
     "tradeoff": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 20,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.6061
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.841
    ],
    "easy-intent-01": [
     "c",
     0.609
    ],
    "easy-intent-02": [
     "c",
     0.583
    ],
    "easy-intent-03": [
     "c",
     0.388
    ],
    "easy-intent-04": [
     "c",
     0.55
    ],
    "easy-intent-05": [
     "c",
     0.589
    ],
    "easy-intent-06": [
     "c",
     0.427
    ],
    "easy-intent-07": [
     "c",
     0.371
    ],
    "easy-intent-08": [
     "c",
     1.083
    ],
    "easy-intent-09": [
     "c",
     0.344
    ],
    "easy-intent-10": [
     "c",
     0.349
    ],
    "easy-intent-11": [
     "c",
     0.36
    ],
    "easy-fact-00": [
     "c",
     0.378
    ],
    "easy-fact-01": [
     "c",
     0.547
    ],
    "easy-fact-02": [
     "c",
     0.425
    ],
    "easy-fact-03": [
     "c",
     0.922
    ],
    "easy-fact-04": [
     "c",
     0.517
    ],
    "easy-fact-05": [
     "c",
     0.42
    ],
    "easy-fact-06": [
     "c",
     0.764
    ],
    "easy-fact-07": [
     "c",
     0.964
    ],
    "easy-fact-08": [
     "c",
     1.127
    ],
    "easy-fact-09": [
     "c",
     0.401
    ],
    "easy-fact-10": [
     "c",
     0.542
    ],
    "easy-fact-11": [
     "c",
     0.598
    ],
    "easy-extraction-00": [
     "c",
     0.899
    ],
    "easy-extraction-01": [
     "c",
     0.584
    ],
    "easy-extraction-02": [
     "c",
     0.769
    ],
    "easy-extraction-03": [
     "c",
     0.479
    ],
    "easy-extraction-04": [
     "c",
     0.525
    ],
    "easy-extraction-05": [
     "c",
     0.54
    ],
    "easy-extraction-06": [
     "c",
     0.548
    ],
    "easy-extraction-07": [
     "c",
     0.583
    ],
    "easy-extraction-08": [
     "c",
     0.59
    ],
    "easy-extraction-09": [
     "c",
     0.579
    ],
    "easy-extraction-10": [
     "c",
     0.594
    ],
    "easy-extraction-11": [
     "c",
     0.583
    ],
    "easy-tool_selection-00": [
     "c",
     0.588
    ],
    "easy-tool_selection-01": [
     "c",
     0.589
    ],
    "easy-tool_selection-02": [
     "c",
     0.556
    ],
    "easy-tool_selection-03": [
     "c",
     0.527
    ],
    "easy-tool_selection-04": [
     "c",
     0.758
    ],
    "easy-tool_selection-05": [
     "c",
     0.949
    ],
    "easy-tool_selection-06": [
     "c",
     0.577
    ],
    "easy-tool_selection-07": [
     "c",
     0.593
    ],
    "easy-tool_selection-08": [
     "c",
     0.58
    ],
    "easy-tool_selection-09": [
     "c",
     0.954
    ],
    "easy-tool_selection-10": [
     "c",
     0.571
    ],
    "easy-tool_selection-11": [
     "c",
     0.779
    ],
    "original-policy-01-0": [
     "c",
     0.48
    ],
    "original-policy-01-1": [
     "c",
     0.357
    ],
    "original-policy-02-0": [
     "c",
     0.365
    ],
    "original-policy-02-1": [
     "c",
     0.382
    ],
    "original-policy-03-0": [
     "c",
     0.366
    ],
    "original-policy-03-1": [
     "c",
     0.523
    ],
    "original-policy-04-0": [
     "c",
     0.593
    ],
    "original-policy-04-1": [
     "c",
     0.981
    ],
    "original-policy-05-0": [
     "c",
     0.75
    ],
    "original-policy-05-1": [
     "c",
     0.577
    ],
    "original-policy-06-0": [
     "c",
     0.893
    ],
    "original-policy-06-1": [
     "c",
     0.912
    ],
    "original-intent-01-0": [
     "c",
     0.583
    ],
    "original-intent-01-1": [
     "c",
     0.605
    ],
    "original-intent-02-0": [
     "w",
     0.586
    ],
    "original-intent-02-1": [
     "c",
     0.956
    ],
    "original-intent-03-0": [
     "c",
     0.599
    ],
    "original-intent-03-1": [
     "c",
     0.602
    ],
    "original-intent-04-0": [
     "c",
     0.602
    ],
    "original-intent-04-1": [
     "c",
     0.575
    ],
    "original-intent-05-0": [
     "c",
     0.585
    ],
    "original-intent-05-1": [
     "c",
     0.571
    ],
    "original-intent-06-0": [
     "c",
     0.575
    ],
    "original-intent-06-1": [
     "c",
     0.591
    ],
    "original-ordinal-01-0": [
     "c",
     0.574
    ],
    "original-ordinal-01-1": [
     "c",
     0.58
    ],
    "original-ordinal-02-0": [
     "c",
     0.925
    ],
    "original-ordinal-02-1": [
     "c",
     0.587
    ],
    "original-ordinal-03-0": [
     "c",
     0.584
    ],
    "original-ordinal-03-1": [
     "c",
     0.583
    ],
    "original-ordinal-04-0": [
     "c",
     0.936
    ],
    "original-ordinal-04-1": [
     "c",
     0.59
    ],
    "original-ordinal-05-0": [
     "c",
     0.573
    ],
    "original-ordinal-05-1": [
     "c",
     0.311
    ],
    "original-ordinal-06-0": [
     "c",
     0.311
    ],
    "original-ordinal-06-1": [
     "c",
     0.355
    ],
    "original-extraction-01-0": [
     "c",
     0.518
    ],
    "original-extraction-01-1": [
     "c",
     0.588
    ],
    "original-extraction-02-0": [
     "c",
     0.606
    ],
    "original-extraction-02-1": [
     "c",
     0.565
    ],
    "original-extraction-03-0": [
     "c",
     0.59
    ],
    "original-extraction-03-1": [
     "c",
     0.593
    ],
    "original-extraction-04-0": [
     "c",
     0.565
    ],
    "original-extraction-04-1": [
     "c",
     0.557
    ],
    "original-extraction-05-0": [
     "c",
     0.612
    ],
    "original-extraction-05-1": [
     "c",
     0.581
    ],
    "original-extraction-06-0": [
     "c",
     0.572
    ],
    "original-extraction-06-1": [
     "c",
     0.583
    ],
    "original-adequacy-01-0": [
     "c",
     0.585
    ],
    "original-adequacy-01-1": [
     "c",
     0.564
    ],
    "original-adequacy-02-0": [
     "w",
     0.557
    ],
    "original-adequacy-02-1": [
     "c",
     0.564
    ],
    "original-adequacy-03-0": [
     "w",
     0.594
    ],
    "original-adequacy-03-1": [
     "w",
     0.569
    ],
    "original-adequacy-04-0": [
     "c",
     0.574
    ],
    "original-adequacy-04-1": [
     "c",
     0.576
    ],
    "original-adequacy-05-0": [
     "w",
     0.549
    ],
    "original-adequacy-05-1": [
     "c",
     0.563
    ],
    "original-adequacy-06-0": [
     "c",
     0.479
    ],
    "original-adequacy-06-1": [
     "c",
     0.312
    ],
    "original-routing-01-0": [
     "c",
     0.716
    ],
    "original-routing-01-1": [
     "c",
     0.532
    ],
    "original-routing-02-0": [
     "c",
     0.75
    ],
    "original-routing-02-1": [
     "c",
     0.545
    ],
    "original-routing-03-0": [
     "c",
     0.59
    ],
    "original-routing-03-1": [
     "c",
     0.808
    ],
    "original-routing-04-0": [
     "c",
     0.566
    ],
    "original-routing-04-1": [
     "c",
     0.586
    ],
    "original-routing-05-0": [
     "c",
     0.921
    ],
    "original-routing-05-1": [
     "c",
     0.566
    ],
    "original-routing-06-0": [
     "c",
     0.575
    ],
    "original-routing-06-1": [
     "c",
     0.558
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     1.925
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.884
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     1.498
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     2.333
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.643
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     1.352
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     1.178
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     1.89
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.467
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.959
    ],
    "hard-opus-a-probability-07": [
     "c",
     0.676
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.977
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.956
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.479
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.663
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.47
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.5
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.613
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.464
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.468
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     1.0
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.465
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.784
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.517
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     1.93
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     1.243
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     1.517
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     1.36
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     1.183
    ],
    "hard-opus-b-probability-01": [
     "w",
     1.303
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.619
    ],
    "hard-opus-b-probability-03": [
     "w",
     0.642
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.474
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.757
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.717
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.565
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.718
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.535
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.827
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     1.22
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     2.082
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     2.166
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.661
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.656
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.51
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     1.604
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     2.584
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.971
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.816
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     1.303
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.241
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.863
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     1.177
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.795
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     0.621
    ],
    "hard-sol-a-adversarial-06": [
     "w",
     0.631
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.868
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.452
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.713
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.425
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     1.437
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     1.281
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     1.727
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     1.95
    ],
    "hard-sol-a-multi_hop-08": [
     "c",
     1.39
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     1.99
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     1.464
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     1.47
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.631
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.553
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.613
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.648
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.519
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.908
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.591
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.629
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.526
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.489
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.671
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.492
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.53
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.526
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.486
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.57
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     2.542
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     2.747
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.684
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     2.009
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.656
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.621
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     1.088
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.428
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.289
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.531
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.827
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.392
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.408
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.783
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     1.004
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.385
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.45
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.775
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.66
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.418
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.429
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.406
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     1.57
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     2.546
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     1.613
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.577
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.62
    ]
   }
  },
  "laya": {
   "display": "Laya (Convai Innovations, ModernBERT-large 421M)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 68,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9444
    },
    "standard": {
     "c": 70,
     "w": 26,
     "f": 0,
     "n": 0,
     "accuracy": 0.7292
    },
    "judge": {
     "c": 101,
     "w": 45,
     "f": 0,
     "n": 0,
     "accuracy": 0.6918
    },
    "hard": {
     "c": 75,
     "w": 145,
     "f": 0,
     "n": 0,
     "accuracy": 0.3409
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 14,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.7778
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "intent": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "ordinal": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "extraction": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "adequacy": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "routing": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     }
    },
    "judge": {
     "routing": {
      "c": 61,
      "w": 17,
      "f": 0,
      "n": 0,
      "accuracy": 0.7821
     },
     "adequacy": {
      "c": 40,
      "w": 28,
      "f": 0,
      "n": 0,
      "accuracy": 0.5882
     }
    },
    "hard": {
     "long_policy": {
      "c": 12,
      "w": 26,
      "f": 0,
      "n": 0,
      "accuracy": 0.3158
     },
     "probability": {
      "c": 8,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 2,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.1429
     },
     "multi_hop": {
      "c": 12,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.3429
     },
     "tradeoff": {
      "c": 6,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 3,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.1875
     },
     "judge_hard": {
      "c": 12,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3636
     },
     "routing_hard": {
      "c": 3,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.508
    ],
    "easy-intent-01": [
     "c",
     0.458
    ],
    "easy-intent-02": [
     "c",
     0.422
    ],
    "easy-intent-03": [
     "c",
     0.39
    ],
    "easy-intent-04": [
     "c",
     0.445
    ],
    "easy-intent-05": [
     "c",
     0.422
    ],
    "easy-intent-06": [
     "c",
     0.388
    ],
    "easy-intent-07": [
     "c",
     0.335
    ],
    "easy-intent-08": [
     "c",
     0.345
    ],
    "easy-intent-09": [
     "c",
     0.368
    ],
    "easy-intent-10": [
     "c",
     0.402
    ],
    "easy-intent-11": [
     "c",
     0.376
    ],
    "easy-fact-00": [
     "c",
     0.361
    ],
    "easy-fact-01": [
     "w",
     0.423
    ],
    "easy-fact-02": [
     "c",
     0.44
    ],
    "easy-fact-03": [
     "c",
     0.35
    ],
    "easy-fact-04": [
     "c",
     0.29
    ],
    "easy-fact-05": [
     "c",
     0.308
    ],
    "easy-fact-06": [
     "w",
     0.369
    ],
    "easy-fact-07": [
     "c",
     0.314
    ],
    "easy-fact-08": [
     "c",
     0.313
    ],
    "easy-fact-09": [
     "c",
     0.3
    ],
    "easy-fact-10": [
     "c",
     0.32
    ],
    "easy-fact-11": [
     "c",
     0.356
    ],
    "easy-extraction-00": [
     "c",
     0.353
    ],
    "easy-extraction-01": [
     "c",
     0.321
    ],
    "easy-extraction-02": [
     "c",
     0.341
    ],
    "easy-extraction-03": [
     "c",
     0.285
    ],
    "easy-extraction-04": [
     "c",
     0.263
    ],
    "easy-extraction-05": [
     "c",
     0.267
    ],
    "easy-extraction-06": [
     "c",
     0.267
    ],
    "easy-extraction-07": [
     "c",
     0.247
    ],
    "easy-extraction-08": [
     "c",
     0.273
    ],
    "easy-extraction-09": [
     "c",
     0.33
    ],
    "easy-extraction-10": [
     "c",
     0.256
    ],
    "easy-extraction-11": [
     "c",
     0.271
    ],
    "easy-tool_selection-00": [
     "c",
     0.38
    ],
    "easy-tool_selection-01": [
     "c",
     0.383
    ],
    "easy-tool_selection-02": [
     "c",
     0.396
    ],
    "easy-tool_selection-03": [
     "c",
     0.375
    ],
    "easy-tool_selection-04": [
     "c",
     0.445
    ],
    "easy-tool_selection-05": [
     "c",
     0.46
    ],
    "easy-tool_selection-06": [
     "c",
     0.446
    ],
    "easy-tool_selection-07": [
     "c",
     0.409
    ],
    "easy-tool_selection-08": [
     "c",
     0.395
    ],
    "easy-tool_selection-09": [
     "c",
     0.448
    ],
    "easy-tool_selection-10": [
     "c",
     0.471
    ],
    "easy-tool_selection-11": [
     "c",
     0.587
    ],
    "original-policy-01-0": [
     "c",
     0.392
    ],
    "original-policy-01-1": [
     "c",
     0.463
    ],
    "original-policy-02-0": [
     "c",
     0.419
    ],
    "original-policy-02-1": [
     "c",
     0.382
    ],
    "original-policy-03-0": [
     "w",
     0.508
    ],
    "original-policy-03-1": [
     "c",
     0.373
    ],
    "original-policy-04-0": [
     "w",
     0.431
    ],
    "original-policy-04-1": [
     "c",
     0.415
    ],
    "original-policy-05-0": [
     "w",
     0.51
    ],
    "original-policy-05-1": [
     "c",
     0.503
    ],
    "original-policy-06-0": [
     "c",
     0.545
    ],
    "original-policy-06-1": [
     "c",
     0.464
    ],
    "original-intent-01-0": [
     "c",
     0.484
    ],
    "original-intent-01-1": [
     "c",
     0.392
    ],
    "original-intent-02-0": [
     "w",
     0.467
    ],
    "original-intent-02-1": [
     "c",
     0.438
    ],
    "original-intent-03-0": [
     "w",
     0.788
    ],
    "original-intent-03-1": [
     "c",
     0.412
    ],
    "original-intent-04-0": [
     "c",
     0.38
    ],
    "original-intent-04-1": [
     "c",
     0.355
    ],
    "original-intent-05-0": [
     "w",
     0.355
    ],
    "original-intent-05-1": [
     "w",
     0.321
    ],
    "original-intent-06-0": [
     "c",
     0.308
    ],
    "original-intent-06-1": [
     "c",
     0.338
    ],
    "original-ordinal-01-0": [
     "c",
     0.37
    ],
    "original-ordinal-01-1": [
     "c",
     0.405
    ],
    "original-ordinal-02-0": [
     "c",
     0.377
    ],
    "original-ordinal-02-1": [
     "c",
     0.4
    ],
    "original-ordinal-03-0": [
     "c",
     0.438
    ],
    "original-ordinal-03-1": [
     "c",
     0.382
    ],
    "original-ordinal-04-0": [
     "c",
     0.381
    ],
    "original-ordinal-04-1": [
     "c",
     0.423
    ],
    "original-ordinal-05-0": [
     "c",
     0.397
    ],
    "original-ordinal-05-1": [
     "w",
     0.381
    ],
    "original-ordinal-06-0": [
     "c",
     0.4
    ],
    "original-ordinal-06-1": [
     "c",
     0.42
    ],
    "original-extraction-01-0": [
     "c",
     0.289
    ],
    "original-extraction-01-1": [
     "c",
     0.316
    ],
    "original-extraction-02-0": [
     "c",
     0.312
    ],
    "original-extraction-02-1": [
     "w",
     0.308
    ],
    "original-extraction-03-0": [
     "w",
     0.316
    ],
    "original-extraction-03-1": [
     "c",
     0.322
    ],
    "original-extraction-04-0": [
     "c",
     0.304
    ],
    "original-extraction-04-1": [
     "c",
     0.344
    ],
    "original-extraction-05-0": [
     "w",
     0.324
    ],
    "original-extraction-05-1": [
     "w",
     0.294
    ],
    "original-extraction-06-0": [
     "c",
     0.319
    ],
    "original-extraction-06-1": [
     "c",
     0.308
    ],
    "original-adequacy-01-0": [
     "w",
     0.281
    ],
    "original-adequacy-01-1": [
     "c",
     0.267
    ],
    "original-adequacy-02-0": [
     "w",
     0.293
    ],
    "original-adequacy-02-1": [
     "c",
     0.294
    ],
    "original-adequacy-03-0": [
     "w",
     0.298
    ],
    "original-adequacy-03-1": [
     "c",
     0.294
    ],
    "original-adequacy-04-0": [
     "c",
     0.322
    ],
    "original-adequacy-04-1": [
     "c",
     0.292
    ],
    "original-adequacy-05-0": [
     "c",
     0.308
    ],
    "original-adequacy-05-1": [
     "c",
     0.284
    ],
    "original-adequacy-06-0": [
     "c",
     0.382
    ],
    "original-adequacy-06-1": [
     "w",
     0.332
    ],
    "original-routing-01-0": [
     "c",
     0.438
    ],
    "original-routing-01-1": [
     "c",
     0.421
    ],
    "original-routing-02-0": [
     "c",
     0.44
    ],
    "original-routing-02-1": [
     "c",
     0.444
    ],
    "original-routing-03-0": [
     "w",
     0.441
    ],
    "original-routing-03-1": [
     "w",
     0.461
    ],
    "original-routing-04-0": [
     "w",
     0.423
    ],
    "original-routing-04-1": [
     "w",
     0.42
    ],
    "original-routing-05-0": [
     "w",
     0.425
    ],
    "original-routing-05-1": [
     "w",
     0.427
    ],
    "original-routing-06-0": [
     "c",
     0.417
    ],
    "original-routing-06-1": [
     "c",
     0.427
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     2.001
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.966
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     2.152
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     2.077
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.975
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     1.992
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     2.033
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     2.056
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.443
    ],
    "hard-opus-a-probability-04": [
     "w",
     1.927
    ],
    "hard-opus-a-probability-07": [
     "c",
     1.981
    ],
    "hard-opus-a-probability-08": [
     "w",
     1.444
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     1.66
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     1.715
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     2.245
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     1.816
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     1.864
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     1.922
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     1.673
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     1.804
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     1.944
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     1.798
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     1.937
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     2.174
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     2.058
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     1.953
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     2.143
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     1.917
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     2.154
    ],
    "hard-opus-b-probability-01": [
     "c",
     2.05
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.745
    ],
    "hard-opus-b-probability-03": [
     "c",
     1.963
    ],
    "hard-opus-b-probability-04": [
     "w",
     1.822
    ],
    "hard-opus-b-probability-06": [
     "c",
     1.945
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     2.09
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     2.115
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     2.644
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     2.629
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     2.289
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     2.106
    ],
    "hard-opus-c-long_policy-02": [
     "w",
     1.904
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     1.922
    ],
    "hard-opus-c-long_policy-04": [
     "w",
     1.946
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.943
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.95
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     2.051
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     2.181
    ],
    "hard-opus-c-probability-03": [
     "w",
     2.083
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     1.997
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     1.963
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.969
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     1.944
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     1.98
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     1.985
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     1.063
    ],
    "hard-sol-a-adversarial-06": [
     "w",
     0.851
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.005
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.782
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.82
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     1.011
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     1.991
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     2.294
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     2.151
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     2.01
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     2.134
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     2.154
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     2.056
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     1.968
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.998
    ],
    "hard-sol-a-trap-04": [
     "w",
     0.693
    ],
    "hard-sol-a-trap-05": [
     "w",
     1.134
    ],
    "hard-sol-a-trap-06": [
     "w",
     1.231
    ],
    "hard-sol-a-trap-08": [
     "w",
     1.288
    ],
    "hard-sol-a-trap-10": [
     "w",
     1.021
    ],
    "hard-sol-a-trap-13": [
     "w",
     1.077
    ],
    "hard-sol-a-trap-15": [
     "w",
     1.712
    ],
    "hard-sol-b-judge_hard-01": [
     "w",
     1.069
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.671
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.823
    ],
    "hard-sol-b-judge_hard-08": [
     "c",
     0.658
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.696
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.791
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.633
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.623
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     2.092
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     2.207
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.965
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     1.92
    ],
    "hard-sol-b-routing_hard-01": [
     "w",
     1.171
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     1.073
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     1.154
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.985
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     1.016
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.884
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.994
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.861
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.794
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     1.564
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     1.208
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     1.094
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     1.118
    ],
    "hard-sol-c-judge_hard-09": [
     "w",
     1.131
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     1.141
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     1.22
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     1.056
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.996
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     1.946
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     1.941
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     2.037
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     2.12
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     2.002
    ]
   }
  },
  "litjev": {
   "display": "LitJev (Qwen3.8-27B)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 94,
     "w": 2,
     "f": 0,
     "n": 0,
     "accuracy": 0.9792
    },
    "judge": {
     "c": 129,
     "w": 17,
     "f": 0,
     "n": 0,
     "accuracy": 0.8836
    },
    "hard": {
     "c": 161,
     "w": 59,
     "f": 0,
     "n": 0,
     "accuracy": 0.7318
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "routing": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     }
    },
    "judge": {
     "routing": {
      "c": 70,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8974
     },
     "adequacy": {
      "c": 59,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.8676
     }
    },
    "hard": {
     "long_policy": {
      "c": 26,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.6842
     },
     "probability": {
      "c": 13,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.65
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 12,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8571
     },
     "multi_hop": {
      "c": 29,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.8286
     },
     "tradeoff": {
      "c": 10,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8333
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 24,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.7273
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     2.164
    ],
    "easy-intent-01": [
     "c",
     0.444
    ],
    "easy-intent-02": [
     "c",
     2.056
    ],
    "easy-intent-03": [
     "c",
     2.309
    ],
    "easy-intent-04": [
     "c",
     1.974
    ],
    "easy-intent-05": [
     "c",
     2.301
    ],
    "easy-intent-06": [
     "c",
     2.002
    ],
    "easy-intent-07": [
     "c",
     2.356
    ],
    "easy-intent-08": [
     "c",
     2.215
    ],
    "easy-intent-09": [
     "c",
     1.896
    ],
    "easy-intent-10": [
     "c",
     1.918
    ],
    "easy-intent-11": [
     "c",
     0.421
    ],
    "easy-fact-00": [
     "c",
     1.908
    ],
    "easy-fact-01": [
     "c",
     1.215
    ],
    "easy-fact-02": [
     "c",
     1.917
    ],
    "easy-fact-03": [
     "c",
     2.228
    ],
    "easy-fact-04": [
     "c",
     2.055
    ],
    "easy-fact-05": [
     "c",
     1.433
    ],
    "easy-fact-06": [
     "c",
     2.167
    ],
    "easy-fact-07": [
     "c",
     2.049
    ],
    "easy-fact-08": [
     "c",
     1.238
    ],
    "easy-fact-09": [
     "c",
     0.57
    ],
    "easy-fact-10": [
     "c",
     1.453
    ],
    "easy-fact-11": [
     "c",
     1.918
    ],
    "easy-extraction-00": [
     "c",
     1.476
    ],
    "easy-extraction-01": [
     "c",
     1.483
    ],
    "easy-extraction-02": [
     "c",
     1.222
    ],
    "easy-extraction-03": [
     "c",
     2.094
    ],
    "easy-extraction-04": [
     "c",
     1.495
    ],
    "easy-extraction-05": [
     "c",
     0.421
    ],
    "easy-extraction-06": [
     "c",
     1.423
    ],
    "easy-extraction-07": [
     "c",
     1.412
    ],
    "easy-extraction-08": [
     "c",
     1.338
    ],
    "easy-extraction-09": [
     "c",
     1.204
    ],
    "easy-extraction-10": [
     "c",
     1.209
    ],
    "easy-extraction-11": [
     "c",
     1.415
    ],
    "easy-tool_selection-00": [
     "c",
     1.453
    ],
    "easy-tool_selection-01": [
     "c",
     1.361
    ],
    "easy-tool_selection-02": [
     "c",
     1.576
    ],
    "easy-tool_selection-03": [
     "c",
     1.416
    ],
    "easy-tool_selection-04": [
     "c",
     1.421
    ],
    "easy-tool_selection-05": [
     "c",
     0.421
    ],
    "easy-tool_selection-06": [
     "c",
     1.254
    ],
    "easy-tool_selection-07": [
     "c",
     0.438
    ],
    "easy-tool_selection-08": [
     "c",
     1.427
    ],
    "easy-tool_selection-09": [
     "c",
     1.216
    ],
    "easy-tool_selection-10": [
     "c",
     0.431
    ],
    "easy-tool_selection-11": [
     "c",
     0.572
    ],
    "original-policy-01-0": [
     "c",
     1.934
    ],
    "original-policy-01-1": [
     "c",
     2.252
    ],
    "original-policy-02-0": [
     "c",
     1.456
    ],
    "original-policy-02-1": [
     "c",
     1.352
    ],
    "original-policy-03-0": [
     "c",
     2.094
    ],
    "original-policy-03-1": [
     "c",
     1.204
    ],
    "original-policy-04-0": [
     "c",
     2.251
    ],
    "original-policy-04-1": [
     "c",
     2.032
    ],
    "original-policy-05-0": [
     "c",
     2.292
    ],
    "original-policy-05-1": [
     "c",
     1.411
    ],
    "original-policy-06-0": [
     "c",
     0.424
    ],
    "original-policy-06-1": [
     "c",
     0.571
    ],
    "original-intent-01-0": [
     "c",
     1.23
    ],
    "original-intent-01-1": [
     "c",
     1.438
    ],
    "original-intent-02-0": [
     "c",
     0.58
    ],
    "original-intent-02-1": [
     "c",
     0.425
    ],
    "original-intent-03-0": [
     "c",
     1.206
    ],
    "original-intent-03-1": [
     "c",
     0.578
    ],
    "original-intent-04-0": [
     "c",
     1.555
    ],
    "original-intent-04-1": [
     "c",
     0.428
    ],
    "original-intent-05-0": [
     "c",
     0.561
    ],
    "original-intent-05-1": [
     "c",
     1.455
    ],
    "original-intent-06-0": [
     "c",
     1.216
    ],
    "original-intent-06-1": [
     "c",
     1.575
    ],
    "original-ordinal-01-0": [
     "c",
     1.228
    ],
    "original-ordinal-01-1": [
     "c",
     1.57
    ],
    "original-ordinal-02-0": [
     "c",
     1.24
    ],
    "original-ordinal-02-1": [
     "c",
     1.582
    ],
    "original-ordinal-03-0": [
     "c",
     0.414
    ],
    "original-ordinal-03-1": [
     "c",
     1.551
    ],
    "original-ordinal-04-0": [
     "c",
     0.554
    ],
    "original-ordinal-04-1": [
     "c",
     1.228
    ],
    "original-ordinal-05-0": [
     "c",
     1.37
    ],
    "original-ordinal-05-1": [
     "c",
     0.576
    ],
    "original-ordinal-06-0": [
     "c",
     0.433
    ],
    "original-ordinal-06-1": [
     "c",
     2.164
    ],
    "original-extraction-01-0": [
     "c",
     1.375
    ],
    "original-extraction-01-1": [
     "c",
     1.231
    ],
    "original-extraction-02-0": [
     "c",
     1.451
    ],
    "original-extraction-02-1": [
     "c",
     0.433
    ],
    "original-extraction-03-0": [
     "c",
     1.23
    ],
    "original-extraction-03-1": [
     "c",
     1.427
    ],
    "original-extraction-04-0": [
     "c",
     0.483
    ],
    "original-extraction-04-1": [
     "c",
     0.438
    ],
    "original-extraction-05-0": [
     "c",
     0.417
    ],
    "original-extraction-05-1": [
     "c",
     1.583
    ],
    "original-extraction-06-0": [
     "c",
     0.572
    ],
    "original-extraction-06-1": [
     "c",
     0.43
    ],
    "original-adequacy-01-0": [
     "c",
     1.383
    ],
    "original-adequacy-01-1": [
     "c",
     1.415
    ],
    "original-adequacy-02-0": [
     "c",
     1.502
    ],
    "original-adequacy-02-1": [
     "c",
     1.36
    ],
    "original-adequacy-03-0": [
     "c",
     0.564
    ],
    "original-adequacy-03-1": [
     "c",
     1.339
    ],
    "original-adequacy-04-0": [
     "c",
     0.425
    ],
    "original-adequacy-04-1": [
     "c",
     0.574
    ],
    "original-adequacy-05-0": [
     "c",
     1.587
    ],
    "original-adequacy-05-1": [
     "c",
     1.348
    ],
    "original-adequacy-06-0": [
     "c",
     1.586
    ],
    "original-adequacy-06-1": [
     "c",
     2.28
    ],
    "original-routing-01-0": [
     "c",
     1.397
    ],
    "original-routing-01-1": [
     "c",
     1.375
    ],
    "original-routing-02-0": [
     "c",
     1.429
    ],
    "original-routing-02-1": [
     "c",
     1.562
    ],
    "original-routing-03-0": [
     "c",
     0.441
    ],
    "original-routing-03-1": [
     "c",
     1.32
    ],
    "original-routing-04-0": [
     "c",
     0.453
    ],
    "original-routing-04-1": [
     "w",
     0.418
    ],
    "original-routing-05-0": [
     "c",
     1.42
    ],
    "original-routing-05-1": [
     "c",
     0.465
    ],
    "original-routing-06-0": [
     "c",
     0.431
    ],
    "original-routing-06-1": [
     "c",
     0.518
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     3.243
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     2.555
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     2.835
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     2.74
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     2.515
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     2.65
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     2.633
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     2.695
    ],
    "hard-opus-a-probability-03": [
     "c",
     2.192
    ],
    "hard-opus-a-probability-04": [
     "w",
     2.273
    ],
    "hard-opus-a-probability-07": [
     "c",
     2.392
    ],
    "hard-opus-a-probability-08": [
     "w",
     2.495
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     2.186
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     2.094
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     2.409
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     2.153
    ],
    "hard-opus-a-temporal_numeric-12": [
     "c",
     1.998
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     1.43
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     2.184
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     1.982
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     2.121
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     2.079
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     2.025
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     2.132
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     2.9
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     2.662
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     2.519
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     2.555
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     2.453
    ],
    "hard-opus-b-probability-01": [
     "c",
     2.671
    ],
    "hard-opus-b-probability-02": [
     "w",
     2.348
    ],
    "hard-opus-b-probability-03": [
     "c",
     2.367
    ],
    "hard-opus-b-probability-04": [
     "c",
     2.131
    ],
    "hard-opus-b-probability-06": [
     "c",
     2.314
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     1.386
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     2.151
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     2.024
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     2.014
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     2.211
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     1.445
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     2.973
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     3.146
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     2.822
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     2.891
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     2.67
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     3.091
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     2.94
    ],
    "hard-opus-c-probability-03": [
     "c",
     1.98
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     2.21
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     2.4
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.359
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     2.423
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     2.082
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     1.492
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     2.158
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     2.133
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.456
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     1.236
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     2.138
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     2.32
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     2.754
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     2.322
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     2.54
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     2.099
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     2.608
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     2.656
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     2.776
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     2.444
    ],
    "hard-sol-a-trap-02": [
     "c",
     2.289
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.588
    ],
    "hard-sol-a-trap-05": [
     "c",
     2.332
    ],
    "hard-sol-a-trap-06": [
     "c",
     2.159
    ],
    "hard-sol-a-trap-08": [
     "c",
     2.111
    ],
    "hard-sol-a-trap-10": [
     "c",
     2.173
    ],
    "hard-sol-a-trap-13": [
     "c",
     1.43
    ],
    "hard-sol-a-trap-15": [
     "c",
     2.085
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     2.077
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     2.039
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     2.3
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     2.076
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     2.257
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     1.993
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     2.197
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     2.346
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     2.688
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     3.181
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     2.729
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     2.697
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     2.058
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.665
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     1.648
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     1.627
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.64
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.587
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     1.609
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     2.27
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     2.127
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     1.633
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     1.599
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     2.392
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     2.304
    ],
    "hard-sol-c-judge_hard-09": [
     "w",
     2.152
    ],
    "hard-sol-c-judge_hard-10": [
     "w",
     2.385
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.609
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     2.094
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.388
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     2.745
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     2.89
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     2.755
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     2.165
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     2.952
    ]
   }
  },
  "opendecision": {
   "display": "OpenDecision (ModernBERT-large zero-shot)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 63,
     "w": 9,
     "f": 0,
     "n": 0,
     "accuracy": 0.875
    },
    "standard": {
     "c": 60,
     "w": 36,
     "f": 0,
     "n": 0,
     "accuracy": 0.625
    },
    "judge": {
     "c": 104,
     "w": 42,
     "f": 0,
     "n": 0,
     "accuracy": 0.7123
    },
    "hard": {
     "c": 73,
     "w": 147,
     "f": 0,
     "n": 0,
     "accuracy": 0.3318
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 9,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "ordinal": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "extraction": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "adequacy": {
      "c": 6,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.375
     },
     "routing": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     }
    },
    "judge": {
     "routing": {
      "c": 47,
      "w": 31,
      "f": 0,
      "n": 0,
      "accuracy": 0.6026
     },
     "adequacy": {
      "c": 57,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.8382
     }
    },
    "hard": {
     "long_policy": {
      "c": 11,
      "w": 27,
      "f": 0,
      "n": 0,
      "accuracy": 0.2895
     },
     "probability": {
      "c": 8,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "temporal_numeric": {
      "c": 7,
      "w": 23,
      "f": 0,
      "n": 0,
      "accuracy": 0.2333
     },
     "ambiguous": {
      "c": 2,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.1429
     },
     "multi_hop": {
      "c": 5,
      "w": 30,
      "f": 0,
      "n": 0,
      "accuracy": 0.1429
     },
     "tradeoff": {
      "c": 3,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.25
     },
     "adversarial": {
      "c": 10,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8333
     },
     "trap": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "judge_hard": {
      "c": 16,
      "w": 17,
      "f": 0,
      "n": 0,
      "accuracy": 0.4848
     },
     "routing_hard": {
      "c": 1,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.1
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.492
    ],
    "easy-intent-01": [
     "c",
     0.458
    ],
    "easy-intent-02": [
     "c",
     0.449
    ],
    "easy-intent-03": [
     "c",
     0.373
    ],
    "easy-intent-04": [
     "c",
     0.369
    ],
    "easy-intent-05": [
     "c",
     0.364
    ],
    "easy-intent-06": [
     "c",
     0.363
    ],
    "easy-intent-07": [
     "c",
     0.374
    ],
    "easy-intent-08": [
     "c",
     0.362
    ],
    "easy-intent-09": [
     "c",
     0.362
    ],
    "easy-intent-10": [
     "c",
     0.36
    ],
    "easy-intent-11": [
     "c",
     0.36
    ],
    "easy-fact-00": [
     "c",
     0.27
    ],
    "easy-fact-01": [
     "w",
     0.268
    ],
    "easy-fact-02": [
     "c",
     0.27
    ],
    "easy-fact-03": [
     "w",
     0.27
    ],
    "easy-fact-04": [
     "c",
     0.268
    ],
    "easy-fact-05": [
     "w",
     0.287
    ],
    "easy-fact-06": [
     "c",
     0.287
    ],
    "easy-fact-07": [
     "w",
     0.287
    ],
    "easy-fact-08": [
     "c",
     0.289
    ],
    "easy-fact-09": [
     "w",
     0.285
    ],
    "easy-fact-10": [
     "c",
     0.287
    ],
    "easy-fact-11": [
     "w",
     0.287
    ],
    "easy-extraction-00": [
     "c",
     0.361
    ],
    "easy-extraction-01": [
     "c",
     0.336
    ],
    "easy-extraction-02": [
     "c",
     0.337
    ],
    "easy-extraction-03": [
     "c",
     0.336
    ],
    "easy-extraction-04": [
     "c",
     0.337
    ],
    "easy-extraction-05": [
     "c",
     0.337
    ],
    "easy-extraction-06": [
     "c",
     0.339
    ],
    "easy-extraction-07": [
     "c",
     0.336
    ],
    "easy-extraction-08": [
     "c",
     0.36
    ],
    "easy-extraction-09": [
     "c",
     0.359
    ],
    "easy-extraction-10": [
     "c",
     0.358
    ],
    "easy-extraction-11": [
     "c",
     0.336
    ],
    "easy-tool_selection-00": [
     "c",
     0.358
    ],
    "easy-tool_selection-01": [
     "c",
     0.358
    ],
    "easy-tool_selection-02": [
     "c",
     0.359
    ],
    "easy-tool_selection-03": [
     "c",
     0.358
    ],
    "easy-tool_selection-04": [
     "c",
     0.36
    ],
    "easy-tool_selection-05": [
     "c",
     0.36
    ],
    "easy-tool_selection-06": [
     "c",
     0.359
    ],
    "easy-tool_selection-07": [
     "c",
     0.359
    ],
    "easy-tool_selection-08": [
     "c",
     0.359
    ],
    "easy-tool_selection-09": [
     "c",
     0.365
    ],
    "easy-tool_selection-10": [
     "c",
     0.361
    ],
    "easy-tool_selection-11": [
     "c",
     0.359
    ],
    "original-policy-01-0": [
     "c",
     0.272
    ],
    "original-policy-01-1": [
     "c",
     0.269
    ],
    "original-policy-02-0": [
     "w",
     0.269
    ],
    "original-policy-02-1": [
     "w",
     0.269
    ],
    "original-policy-03-0": [
     "w",
     0.268
    ],
    "original-policy-03-1": [
     "w",
     0.288
    ],
    "original-policy-04-0": [
     "c",
     0.291
    ],
    "original-policy-04-1": [
     "c",
     0.288
    ],
    "original-policy-05-0": [
     "c",
     0.288
    ],
    "original-policy-05-1": [
     "c",
     0.286
    ],
    "original-policy-06-0": [
     "w",
     0.288
    ],
    "original-policy-06-1": [
     "w",
     0.293
    ],
    "original-intent-01-0": [
     "c",
     0.38
    ],
    "original-intent-01-1": [
     "c",
     0.36
    ],
    "original-intent-02-0": [
     "c",
     0.36
    ],
    "original-intent-02-1": [
     "c",
     0.359
    ],
    "original-intent-03-0": [
     "c",
     0.36
    ],
    "original-intent-03-1": [
     "c",
     0.364
    ],
    "original-intent-04-0": [
     "c",
     0.362
    ],
    "original-intent-04-1": [
     "c",
     0.367
    ],
    "original-intent-05-0": [
     "w",
     0.382
    ],
    "original-intent-05-1": [
     "w",
     0.382
    ],
    "original-intent-06-0": [
     "w",
     0.362
    ],
    "original-intent-06-1": [
     "w",
     0.36
    ],
    "original-ordinal-01-0": [
     "c",
     0.294
    ],
    "original-ordinal-01-1": [
     "c",
     0.295
    ],
    "original-ordinal-02-0": [
     "c",
     0.296
    ],
    "original-ordinal-02-1": [
     "c",
     0.296
    ],
    "original-ordinal-03-0": [
     "c",
     0.293
    ],
    "original-ordinal-03-1": [
     "c",
     0.293
    ],
    "original-ordinal-04-0": [
     "c",
     0.293
    ],
    "original-ordinal-04-1": [
     "c",
     0.295
    ],
    "original-ordinal-05-0": [
     "w",
     0.298
    ],
    "original-ordinal-05-1": [
     "c",
     0.292
    ],
    "original-ordinal-06-0": [
     "c",
     0.297
    ],
    "original-ordinal-06-1": [
     "c",
     0.309
    ],
    "original-extraction-01-0": [
     "w",
     0.338
    ],
    "original-extraction-01-1": [
     "w",
     0.362
    ],
    "original-extraction-02-0": [
     "c",
     0.36
    ],
    "original-extraction-02-1": [
     "w",
     0.362
    ],
    "original-extraction-03-0": [
     "c",
     0.341
    ],
    "original-extraction-03-1": [
     "c",
     0.361
    ],
    "original-extraction-04-0": [
     "w",
     0.363
    ],
    "original-extraction-04-1": [
     "c",
     0.338
    ],
    "original-extraction-05-0": [
     "w",
     0.361
    ],
    "original-extraction-05-1": [
     "w",
     0.36
    ],
    "original-extraction-06-0": [
     "c",
     0.34
    ],
    "original-extraction-06-1": [
     "c",
     0.36
    ],
    "original-adequacy-01-0": [
     "w",
     0.269
    ],
    "original-adequacy-01-1": [
     "c",
     0.269
    ],
    "original-adequacy-02-0": [
     "w",
     0.269
    ],
    "original-adequacy-02-1": [
     "c",
     0.269
    ],
    "original-adequacy-03-0": [
     "w",
     0.27
    ],
    "original-adequacy-03-1": [
     "w",
     0.286
    ],
    "original-adequacy-04-0": [
     "c",
     0.286
    ],
    "original-adequacy-04-1": [
     "w",
     0.287
    ],
    "original-adequacy-05-0": [
     "w",
     0.287
    ],
    "original-adequacy-05-1": [
     "w",
     0.288
    ],
    "original-adequacy-06-0": [
     "w",
     0.297
    ],
    "original-adequacy-06-1": [
     "w",
     0.286
    ],
    "original-routing-01-0": [
     "c",
     0.411
    ],
    "original-routing-01-1": [
     "w",
     0.405
    ],
    "original-routing-02-0": [
     "c",
     0.386
    ],
    "original-routing-02-1": [
     "c",
     0.382
    ],
    "original-routing-03-0": [
     "c",
     0.385
    ],
    "original-routing-03-1": [
     "c",
     0.392
    ],
    "original-routing-04-0": [
     "c",
     0.404
    ],
    "original-routing-04-1": [
     "w",
     0.414
    ],
    "original-routing-05-0": [
     "w",
     0.41
    ],
    "original-routing-05-1": [
     "c",
     0.41
    ],
    "original-routing-06-0": [
     "c",
     0.408
    ],
    "original-routing-06-1": [
     "c",
     0.409
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.569
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.4
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     0.398
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.399
    ],
    "hard-opus-a-long_policy-11": [
     "w",
     0.302
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.309
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.425
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.3
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.278
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.33
    ],
    "hard-opus-a-probability-07": [
     "c",
     0.331
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.273
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.278
    ],
    "hard-opus-a-temporal_numeric-06": [
     "w",
     0.283
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     0.41
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.279
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.314
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.375
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.335
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.333
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.329
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.372
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.382
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.373
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.401
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.453
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.326
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.454
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.298
    ],
    "hard-opus-b-probability-01": [
     "c",
     0.363
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.329
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.283
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.378
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.356
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.411
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.381
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     0.282
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.419
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.283
    ],
    "hard-opus-b-tradeoff-12": [
     "w",
     0.421
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.529
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.59
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.441
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.436
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.435
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.499
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     0.452
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.329
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.289
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     0.408
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.408
    ],
    "hard-opus-c-temporal_numeric-06": [
     "c",
     0.435
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     0.409
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     0.329
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.352
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.277
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.364
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.273
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.302
    ],
    "hard-sol-a-adversarial-11": [
     "w",
     0.365
    ],
    "hard-sol-a-multi_hop-01": [
     "w",
     0.399
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.399
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     0.398
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.37
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.369
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     0.371
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.37
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.37
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.273
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.281
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.303
    ],
    "hard-sol-a-trap-06": [
     "w",
     0.358
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.359
    ],
    "hard-sol-a-trap-10": [
     "w",
     0.385
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.272
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.273
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.273
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.271
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.271
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.273
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.276
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.274
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.272
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.274
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     0.53
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     0.495
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     0.495
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.497
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.417
    ],
    "hard-sol-b-routing_hard-02": [
     "w",
     0.404
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     0.378
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.383
    ],
    "hard-sol-b-routing_hard-09": [
     "w",
     0.378
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.362
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.377
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.377
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.331
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.281
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     0.274
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.272
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.273
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.273
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.274
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.272
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.273
    ],
    "hard-sol-c-judge_hard-15": [
     "w",
     0.274
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.401
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     0.4
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.369
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.386
    ],
    "hard-sol-c-multi_hop-13": [
     "w",
     0.383
    ]
   }
  },
  "openjev-verdict-1.4": {
   "display": "openJev Verdict 1.4",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 62,
     "w": 10,
     "f": 0,
     "n": 0,
     "accuracy": 0.8611
    },
    "standard": {
     "c": 65,
     "w": 31,
     "f": 0,
     "n": 0,
     "accuracy": 0.6771
    },
    "judge": {
     "c": 82,
     "w": 64,
     "f": 0,
     "n": 0,
     "accuracy": 0.5616
    },
    "hard": {
     "c": 83,
     "w": 137,
     "f": 0,
     "n": 0,
     "accuracy": 0.3773
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 9,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "extraction": {
      "c": 17,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9444
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 12,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "ordinal": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "extraction": {
      "c": 13,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.8125
     },
     "adequacy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "routing": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     }
    },
    "judge": {
     "routing": {
      "c": 21,
      "w": 57,
      "f": 0,
      "n": 0,
      "accuracy": 0.2692
     },
     "adequacy": {
      "c": 61,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8971
     }
    },
    "hard": {
     "long_policy": {
      "c": 11,
      "w": 27,
      "f": 0,
      "n": 0,
      "accuracy": 0.2895
     },
     "probability": {
      "c": 8,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "temporal_numeric": {
      "c": 9,
      "w": 21,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     },
     "ambiguous": {
      "c": 6,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.4286
     },
     "multi_hop": {
      "c": 6,
      "w": 29,
      "f": 0,
      "n": 0,
      "accuracy": 0.1714
     },
     "tradeoff": {
      "c": 5,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.4167
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 9,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5625
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 3,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.167
    ],
    "easy-intent-01": [
     "c",
     0.127
    ],
    "easy-intent-02": [
     "c",
     0.134
    ],
    "easy-intent-03": [
     "c",
     0.123
    ],
    "easy-intent-04": [
     "c",
     0.125
    ],
    "easy-intent-05": [
     "c",
     0.128
    ],
    "easy-intent-06": [
     "c",
     0.117
    ],
    "easy-intent-07": [
     "c",
     0.109
    ],
    "easy-intent-08": [
     "c",
     0.13
    ],
    "easy-intent-09": [
     "c",
     0.141
    ],
    "easy-intent-10": [
     "c",
     0.123
    ],
    "easy-intent-11": [
     "c",
     0.116
    ],
    "easy-fact-00": [
     "c",
     0.21
    ],
    "easy-fact-01": [
     "w",
     0.19
    ],
    "easy-fact-02": [
     "c",
     0.188
    ],
    "easy-fact-03": [
     "w",
     0.193
    ],
    "easy-fact-04": [
     "c",
     0.192
    ],
    "easy-fact-05": [
     "w",
     0.204
    ],
    "easy-fact-06": [
     "c",
     0.237
    ],
    "easy-fact-07": [
     "w",
     0.213
    ],
    "easy-fact-08": [
     "c",
     0.205
    ],
    "easy-fact-09": [
     "w",
     0.195
    ],
    "easy-fact-10": [
     "c",
     0.205
    ],
    "easy-fact-11": [
     "w",
     0.209
    ],
    "easy-extraction-00": [
     "c",
     0.089
    ],
    "easy-extraction-01": [
     "c",
     0.114
    ],
    "easy-extraction-02": [
     "c",
     0.091
    ],
    "easy-extraction-03": [
     "c",
     0.097
    ],
    "easy-extraction-04": [
     "c",
     0.105
    ],
    "easy-extraction-05": [
     "c",
     0.1
    ],
    "easy-extraction-06": [
     "c",
     0.086
    ],
    "easy-extraction-07": [
     "c",
     0.087
    ],
    "easy-extraction-08": [
     "c",
     0.087
    ],
    "easy-extraction-09": [
     "c",
     0.089
    ],
    "easy-extraction-10": [
     "c",
     0.084
    ],
    "easy-extraction-11": [
     "c",
     0.092
    ],
    "easy-tool_selection-00": [
     "c",
     0.121
    ],
    "easy-tool_selection-01": [
     "c",
     0.13
    ],
    "easy-tool_selection-02": [
     "c",
     0.117
    ],
    "easy-tool_selection-03": [
     "c",
     0.122
    ],
    "easy-tool_selection-04": [
     "c",
     0.113
    ],
    "easy-tool_selection-05": [
     "c",
     0.112
    ],
    "easy-tool_selection-06": [
     "c",
     0.121
    ],
    "easy-tool_selection-07": [
     "c",
     0.124
    ],
    "easy-tool_selection-08": [
     "c",
     0.126
    ],
    "easy-tool_selection-09": [
     "c",
     0.115
    ],
    "easy-tool_selection-10": [
     "c",
     0.129
    ],
    "easy-tool_selection-11": [
     "c",
     0.117
    ],
    "original-policy-01-0": [
     "w",
     0.261
    ],
    "original-policy-01-1": [
     "w",
     0.228
    ],
    "original-policy-02-0": [
     "c",
     0.222
    ],
    "original-policy-02-1": [
     "c",
     0.268
    ],
    "original-policy-03-0": [
     "c",
     0.218
    ],
    "original-policy-03-1": [
     "c",
     0.213
    ],
    "original-policy-04-0": [
     "w",
     0.223
    ],
    "original-policy-04-1": [
     "w",
     0.224
    ],
    "original-policy-05-0": [
     "w",
     0.226
    ],
    "original-policy-05-1": [
     "w",
     0.217
    ],
    "original-policy-06-0": [
     "c",
     0.259
    ],
    "original-policy-06-1": [
     "c",
     0.283
    ],
    "original-intent-01-0": [
     "c",
     0.142
    ],
    "original-intent-01-1": [
     "c",
     0.151
    ],
    "original-intent-02-0": [
     "w",
     0.139
    ],
    "original-intent-02-1": [
     "w",
     0.13
    ],
    "original-intent-03-0": [
     "c",
     0.114
    ],
    "original-intent-03-1": [
     "c",
     0.128
    ],
    "original-intent-04-0": [
     "c",
     0.132
    ],
    "original-intent-04-1": [
     "w",
     0.127
    ],
    "original-intent-05-0": [
     "c",
     0.13
    ],
    "original-intent-05-1": [
     "c",
     0.133
    ],
    "original-intent-06-0": [
     "c",
     0.141
    ],
    "original-intent-06-1": [
     "c",
     0.159
    ],
    "original-ordinal-01-0": [
     "c",
     0.229
    ],
    "original-ordinal-01-1": [
     "c",
     0.183
    ],
    "original-ordinal-02-0": [
     "c",
     0.182
    ],
    "original-ordinal-02-1": [
     "c",
     0.184
    ],
    "original-ordinal-03-0": [
     "c",
     0.175
    ],
    "original-ordinal-03-1": [
     "c",
     0.217
    ],
    "original-ordinal-04-0": [
     "c",
     0.242
    ],
    "original-ordinal-04-1": [
     "c",
     0.195
    ],
    "original-ordinal-05-0": [
     "w",
     0.227
    ],
    "original-ordinal-05-1": [
     "c",
     0.199
    ],
    "original-ordinal-06-0": [
     "c",
     0.21
    ],
    "original-ordinal-06-1": [
     "c",
     0.193
    ],
    "original-extraction-01-0": [
     "w",
     0.169
    ],
    "original-extraction-01-1": [
     "w",
     0.142
    ],
    "original-extraction-02-0": [
     "c",
     0.156
    ],
    "original-extraction-02-1": [
     "c",
     0.18
    ],
    "original-extraction-03-0": [
     "c",
     0.162
    ],
    "original-extraction-03-1": [
     "c",
     0.169
    ],
    "original-extraction-04-0": [
     "c",
     0.178
    ],
    "original-extraction-04-1": [
     "c",
     0.159
    ],
    "original-extraction-05-0": [
     "c",
     0.17
    ],
    "original-extraction-05-1": [
     "c",
     0.209
    ],
    "original-extraction-06-0": [
     "c",
     0.141
    ],
    "original-extraction-06-1": [
     "c",
     0.173
    ],
    "original-adequacy-01-0": [
     "c",
     0.248
    ],
    "original-adequacy-01-1": [
     "c",
     0.327
    ],
    "original-adequacy-02-0": [
     "w",
     0.271
    ],
    "original-adequacy-02-1": [
     "w",
     0.272
    ],
    "original-adequacy-03-0": [
     "w",
     0.219
    ],
    "original-adequacy-03-1": [
     "w",
     0.25
    ],
    "original-adequacy-04-0": [
     "c",
     0.276
    ],
    "original-adequacy-04-1": [
     "c",
     0.237
    ],
    "original-adequacy-05-0": [
     "w",
     0.266
    ],
    "original-adequacy-05-1": [
     "w",
     0.226
    ],
    "original-adequacy-06-0": [
     "c",
     0.314
    ],
    "original-adequacy-06-1": [
     "c",
     0.283
    ],
    "original-routing-01-0": [
     "c",
     0.184
    ],
    "original-routing-01-1": [
     "w",
     0.183
    ],
    "original-routing-02-0": [
     "c",
     0.177
    ],
    "original-routing-02-1": [
     "c",
     0.174
    ],
    "original-routing-03-0": [
     "c",
     0.156
    ],
    "original-routing-03-1": [
     "c",
     0.186
    ],
    "original-routing-04-0": [
     "c",
     0.174
    ],
    "original-routing-04-1": [
     "c",
     0.168
    ],
    "original-routing-05-0": [
     "c",
     0.178
    ],
    "original-routing-05-1": [
     "w",
     0.197
    ],
    "original-routing-06-0": [
     "w",
     0.142
    ],
    "original-routing-06-1": [
     "w",
     0.142
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     0.862
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.766
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     0.731
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     0.684
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.908
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     0.84
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     0.753
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     0.838
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.767
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.671
    ],
    "hard-opus-a-probability-07": [
     "w",
     0.633
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.506
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     0.623
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.622
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.61
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.649
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.641
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.628
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.597
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.568
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.616
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.587
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     0.505
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.644
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     0.631
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.619
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     0.756
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     0.709
    ],
    "hard-opus-b-multi_hop-08": [
     "w",
     0.631
    ],
    "hard-opus-b-probability-01": [
     "w",
     0.622
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.569
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.64
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.57
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.815
    ],
    "hard-opus-b-tradeoff-01": [
     "c",
     0.667
    ],
    "hard-opus-b-tradeoff-03": [
     "w",
     0.652
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.609
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.623
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.612
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.633
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     0.641
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.635
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.634
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.613
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     0.634
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.621
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.621
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.62
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     0.764
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.636
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     0.624
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.634
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.683
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.611
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.328
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.374
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     0.301
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.401
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.29
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.342
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.645
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     0.629
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     0.675
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.646
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.625
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     0.63
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.69
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.699
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.569
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.313
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.395
    ],
    "hard-sol-a-trap-06": [
     "w",
     0.368
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.302
    ],
    "hard-sol-a-trap-10": [
     "w",
     0.219
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.328
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.432
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.426
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.484
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.399
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.343
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.37
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.393
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.347
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.344
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     0.705
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     0.694
    ],
    "hard-sol-b-long_policy-05": [
     "w",
     0.772
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     0.709
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.43
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.333
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     0.401
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.399
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.324
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.256
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.273
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     0.215
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.24
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.616
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.547
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.496
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.507
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.498
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.538
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.544
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.534
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.485
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     0.678
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     0.62
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     0.647
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.623
    ],
    "hard-sol-c-multi_hop-13": [
     "w",
     0.657
    ]
   }
  },
  "openjev-verdict": {
   "display": "openJev Verdict (heman10x, ModernBERT-base 151M)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 62,
     "w": 10,
     "f": 0,
     "n": 0,
     "accuracy": 0.8611
    },
    "standard": {
     "c": 63,
     "w": 33,
     "f": 0,
     "n": 0,
     "accuracy": 0.6562
    },
    "judge": {
     "c": 89,
     "w": 57,
     "f": 0,
     "n": 0,
     "accuracy": 0.6096
    },
    "hard": {
     "c": 84,
     "w": 136,
     "f": 0,
     "n": 0,
     "accuracy": 0.3818
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 9,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "extraction": {
      "c": 17,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9444
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "intent": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "ordinal": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "extraction": {
      "c": 10,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.625
     },
     "adequacy": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "routing": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     }
    },
    "judge": {
     "routing": {
      "c": 29,
      "w": 49,
      "f": 0,
      "n": 0,
      "accuracy": 0.3718
     },
     "adequacy": {
      "c": 60,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.8824
     }
    },
    "hard": {
     "long_policy": {
      "c": 11,
      "w": 27,
      "f": 0,
      "n": 0,
      "accuracy": 0.2895
     },
     "probability": {
      "c": 9,
      "w": 11,
      "f": 0,
      "n": 0,
      "accuracy": 0.45
     },
     "temporal_numeric": {
      "c": 12,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "ambiguous": {
      "c": 6,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.4286
     },
     "multi_hop": {
      "c": 5,
      "w": 30,
      "f": 0,
      "n": 0,
      "accuracy": 0.1429
     },
     "tradeoff": {
      "c": 4,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "adversarial": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "trap": {
      "c": 8,
      "w": 8,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "judge_hard": {
      "c": 18,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.5455
     },
     "routing_hard": {
      "c": 3,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.3
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.179
    ],
    "easy-intent-01": [
     "c",
     0.177
    ],
    "easy-intent-02": [
     "c",
     0.16
    ],
    "easy-intent-03": [
     "c",
     0.137
    ],
    "easy-intent-04": [
     "c",
     0.136
    ],
    "easy-intent-05": [
     "c",
     0.136
    ],
    "easy-intent-06": [
     "c",
     0.124
    ],
    "easy-intent-07": [
     "c",
     0.105
    ],
    "easy-intent-08": [
     "c",
     0.101
    ],
    "easy-intent-09": [
     "c",
     0.107
    ],
    "easy-intent-10": [
     "c",
     0.108
    ],
    "easy-intent-11": [
     "c",
     0.105
    ],
    "easy-fact-00": [
     "c",
     0.225
    ],
    "easy-fact-01": [
     "w",
     0.19
    ],
    "easy-fact-02": [
     "c",
     0.205
    ],
    "easy-fact-03": [
     "w",
     0.203
    ],
    "easy-fact-04": [
     "c",
     0.205
    ],
    "easy-fact-05": [
     "w",
     0.202
    ],
    "easy-fact-06": [
     "c",
     0.28
    ],
    "easy-fact-07": [
     "w",
     0.234
    ],
    "easy-fact-08": [
     "c",
     0.233
    ],
    "easy-fact-09": [
     "w",
     0.28
    ],
    "easy-fact-10": [
     "c",
     0.249
    ],
    "easy-fact-11": [
     "w",
     0.209
    ],
    "easy-extraction-00": [
     "c",
     0.079
    ],
    "easy-extraction-01": [
     "c",
     0.123
    ],
    "easy-extraction-02": [
     "c",
     0.11
    ],
    "easy-extraction-03": [
     "c",
     0.116
    ],
    "easy-extraction-04": [
     "c",
     0.11
    ],
    "easy-extraction-05": [
     "c",
     0.106
    ],
    "easy-extraction-06": [
     "w",
     0.111
    ],
    "easy-extraction-07": [
     "c",
     0.11
    ],
    "easy-extraction-08": [
     "c",
     0.084
    ],
    "easy-extraction-09": [
     "c",
     0.091
    ],
    "easy-extraction-10": [
     "c",
     0.09
    ],
    "easy-extraction-11": [
     "c",
     0.099
    ],
    "easy-tool_selection-00": [
     "c",
     0.114
    ],
    "easy-tool_selection-01": [
     "c",
     0.114
    ],
    "easy-tool_selection-02": [
     "c",
     0.115
    ],
    "easy-tool_selection-03": [
     "c",
     0.134
    ],
    "easy-tool_selection-04": [
     "c",
     0.14
    ],
    "easy-tool_selection-05": [
     "c",
     0.143
    ],
    "easy-tool_selection-06": [
     "c",
     0.149
    ],
    "easy-tool_selection-07": [
     "c",
     0.169
    ],
    "easy-tool_selection-08": [
     "c",
     0.112
    ],
    "easy-tool_selection-09": [
     "c",
     0.108
    ],
    "easy-tool_selection-10": [
     "c",
     0.114
    ],
    "easy-tool_selection-11": [
     "c",
     0.122
    ],
    "original-policy-01-0": [
     "w",
     0.258
    ],
    "original-policy-01-1": [
     "w",
     0.263
    ],
    "original-policy-02-0": [
     "c",
     0.215
    ],
    "original-policy-02-1": [
     "c",
     0.219
    ],
    "original-policy-03-0": [
     "c",
     0.224
    ],
    "original-policy-03-1": [
     "c",
     0.208
    ],
    "original-policy-04-0": [
     "w",
     0.215
    ],
    "original-policy-04-1": [
     "w",
     0.22
    ],
    "original-policy-05-0": [
     "w",
     0.217
    ],
    "original-policy-05-1": [
     "w",
     0.214
    ],
    "original-policy-06-0": [
     "c",
     0.217
    ],
    "original-policy-06-1": [
     "c",
     0.202
    ],
    "original-intent-01-0": [
     "c",
     0.103
    ],
    "original-intent-01-1": [
     "w",
     0.101
    ],
    "original-intent-02-0": [
     "w",
     0.149
    ],
    "original-intent-02-1": [
     "w",
     0.114
    ],
    "original-intent-03-0": [
     "c",
     0.107
    ],
    "original-intent-03-1": [
     "c",
     0.11
    ],
    "original-intent-04-0": [
     "c",
     0.132
    ],
    "original-intent-04-1": [
     "w",
     0.106
    ],
    "original-intent-05-0": [
     "c",
     0.109
    ],
    "original-intent-05-1": [
     "c",
     0.117
    ],
    "original-intent-06-0": [
     "c",
     0.114
    ],
    "original-intent-06-1": [
     "w",
     0.119
    ],
    "original-ordinal-01-0": [
     "c",
     0.16
    ],
    "original-ordinal-01-1": [
     "c",
     0.152
    ],
    "original-ordinal-02-0": [
     "c",
     0.158
    ],
    "original-ordinal-02-1": [
     "c",
     0.156
    ],
    "original-ordinal-03-0": [
     "c",
     0.179
    ],
    "original-ordinal-03-1": [
     "c",
     0.152
    ],
    "original-ordinal-04-0": [
     "c",
     0.153
    ],
    "original-ordinal-04-1": [
     "c",
     0.148
    ],
    "original-ordinal-05-0": [
     "w",
     0.145
    ],
    "original-ordinal-05-1": [
     "c",
     0.147
    ],
    "original-ordinal-06-0": [
     "c",
     0.145
    ],
    "original-ordinal-06-1": [
     "c",
     0.168
    ],
    "original-extraction-01-0": [
     "w",
     0.107
    ],
    "original-extraction-01-1": [
     "w",
     0.106
    ],
    "original-extraction-02-0": [
     "c",
     0.098
    ],
    "original-extraction-02-1": [
     "c",
     0.099
    ],
    "original-extraction-03-0": [
     "w",
     0.119
    ],
    "original-extraction-03-1": [
     "c",
     0.114
    ],
    "original-extraction-04-0": [
     "w",
     0.1
    ],
    "original-extraction-04-1": [
     "w",
     0.102
    ],
    "original-extraction-05-0": [
     "c",
     0.098
    ],
    "original-extraction-05-1": [
     "c",
     0.094
    ],
    "original-extraction-06-0": [
     "c",
     0.114
    ],
    "original-extraction-06-1": [
     "c",
     0.118
    ],
    "original-adequacy-01-0": [
     "c",
     0.21
    ],
    "original-adequacy-01-1": [
     "c",
     0.185
    ],
    "original-adequacy-02-0": [
     "w",
     0.197
    ],
    "original-adequacy-02-1": [
     "w",
     0.196
    ],
    "original-adequacy-03-0": [
     "w",
     0.193
    ],
    "original-adequacy-03-1": [
     "w",
     0.198
    ],
    "original-adequacy-04-0": [
     "c",
     0.225
    ],
    "original-adequacy-04-1": [
     "c",
     0.216
    ],
    "original-adequacy-05-0": [
     "w",
     0.197
    ],
    "original-adequacy-05-1": [
     "w",
     0.185
    ],
    "original-adequacy-06-0": [
     "c",
     0.201
    ],
    "original-adequacy-06-1": [
     "c",
     0.204
    ],
    "original-routing-01-0": [
     "c",
     0.137
    ],
    "original-routing-01-1": [
     "w",
     0.126
    ],
    "original-routing-02-0": [
     "c",
     0.135
    ],
    "original-routing-02-1": [
     "c",
     0.13
    ],
    "original-routing-03-0": [
     "c",
     0.133
    ],
    "original-routing-03-1": [
     "c",
     0.143
    ],
    "original-routing-04-0": [
     "c",
     0.123
    ],
    "original-routing-04-1": [
     "c",
     0.137
    ],
    "original-routing-05-0": [
     "c",
     0.141
    ],
    "original-routing-05-1": [
     "w",
     0.122
    ],
    "original-routing-06-0": [
     "w",
     0.121
    ],
    "original-routing-06-1": [
     "w",
     0.123
    ],
    "hard-opus-a-long_policy-01": [
     "w",
     1.879
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     1.63
    ],
    "hard-opus-a-long_policy-08": [
     "w",
     1.548
    ],
    "hard-opus-a-long_policy-09": [
     "w",
     1.546
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.548
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     1.465
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     1.5
    ],
    "hard-opus-a-long_policy-19": [
     "c",
     1.397
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.62
    ],
    "hard-opus-a-probability-04": [
     "w",
     1.22
    ],
    "hard-opus-a-probability-07": [
     "w",
     1.568
    ],
    "hard-opus-a-probability-08": [
     "w",
     0.601
    ],
    "hard-opus-a-temporal_numeric-03": [
     "c",
     0.841
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.856
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     1.407
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     1.408
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.706
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     0.67
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.673
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     0.683
    ],
    "hard-opus-b-ambiguous-09": [
     "w",
     0.845
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     0.534
    ],
    "hard-opus-b-ambiguous-11": [
     "w",
     0.48
    ],
    "hard-opus-b-ambiguous-13": [
     "w",
     0.675
    ],
    "hard-opus-b-multi_hop-03": [
     "w",
     1.554
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     1.45
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     1.466
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     1.436
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     1.424
    ],
    "hard-opus-b-probability-01": [
     "c",
     1.358
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.529
    ],
    "hard-opus-b-probability-03": [
     "c",
     1.437
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.538
    ],
    "hard-opus-b-probability-06": [
     "c",
     1.126
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.975
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.785
    ],
    "hard-opus-b-tradeoff-06": [
     "w",
     1.073
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.808
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.84
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.623
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     1.91
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     1.585
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.547
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.374
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     1.335
    ],
    "hard-opus-c-long_policy-10": [
     "w",
     1.473
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     1.379
    ],
    "hard-opus-c-probability-03": [
     "w",
     0.665
    ],
    "hard-opus-c-temporal_numeric-02": [
     "c",
     1.382
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.989
    ],
    "hard-opus-c-temporal_numeric-04": [
     "c",
     0.968
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     1.398
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     1.264
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     1.057
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.369
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.462
    ],
    "hard-sol-a-adversarial-07": [
     "w",
     0.425
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.472
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.35
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.373
    ],
    "hard-sol-a-multi_hop-01": [
     "w",
     1.638
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     1.429
    ],
    "hard-sol-a-multi_hop-05": [
     "w",
     1.427
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     1.405
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     1.488
    ],
    "hard-sol-a-multi_hop-09": [
     "w",
     1.542
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     1.5
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     1.438
    ],
    "hard-sol-a-trap-02": [
     "w",
     0.469
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.336
    ],
    "hard-sol-a-trap-05": [
     "w",
     0.378
    ],
    "hard-sol-a-trap-06": [
     "w",
     0.35
    ],
    "hard-sol-a-trap-08": [
     "w",
     0.266
    ],
    "hard-sol-a-trap-10": [
     "w",
     0.194
    ],
    "hard-sol-a-trap-13": [
     "w",
     0.418
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.431
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.385
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.338
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.391
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.319
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.373
    ],
    "hard-sol-b-judge_hard-14": [
     "w",
     0.375
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.317
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.303
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     1.443
    ],
    "hard-sol-b-long_policy-02": [
     "w",
     1.869
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.552
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     1.478
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.37
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.314
    ],
    "hard-sol-b-routing_hard-03": [
     "w",
     0.374
    ],
    "hard-sol-b-routing_hard-07": [
     "w",
     0.298
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.297
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.26
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.249
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.214
    ],
    "hard-sol-b-temporal_numeric-04": [
     "c",
     0.231
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     0.655
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.617
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.521
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     0.502
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.505
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.55
    ],
    "hard-sol-c-judge_hard-11": [
     "w",
     0.575
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.509
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.485
    ],
    "hard-sol-c-multi_hop-07": [
     "w",
     1.406
    ],
    "hard-sol-c-multi_hop-09": [
     "w",
     1.414
    ],
    "hard-sol-c-multi_hop-11": [
     "w",
     1.611
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     0.575
    ],
    "hard-sol-c-multi_hop-13": [
     "w",
     1.387
    ]
   }
  },
  "reflex-27b": {
   "display": "reflex-27b (Qwen3.8-27B)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 92,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9583
    },
    "judge": {
     "c": 140,
     "w": 6,
     "f": 0,
     "n": 0,
     "accuracy": 0.9589
    },
    "hard": {
     "c": 167,
     "w": 53,
     "f": 0,
     "n": 0,
     "accuracy": 0.7591
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 77,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9872
     },
     "adequacy": {
      "c": 63,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.9265
     }
    },
    "hard": {
     "long_policy": {
      "c": 26,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.6842
     },
     "probability": {
      "c": 13,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.65
     },
     "temporal_numeric": {
      "c": 12,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.4
     },
     "ambiguous": {
      "c": 13,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9286
     },
     "multi_hop": {
      "c": 28,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.8
     },
     "tradeoff": {
      "c": 10,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8333
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 27,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.8182
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     33.115
    ],
    "easy-intent-01": [
     "c",
     0.412
    ],
    "easy-intent-02": [
     "c",
     2.195
    ],
    "easy-intent-03": [
     "c",
     1.971
    ],
    "easy-intent-04": [
     "c",
     2.169
    ],
    "easy-intent-05": [
     "c",
     2.044
    ],
    "easy-intent-06": [
     "c",
     1.986
    ],
    "easy-intent-07": [
     "c",
     2.171
    ],
    "easy-intent-08": [
     "c",
     2.118
    ],
    "easy-intent-09": [
     "c",
     1.903
    ],
    "easy-intent-10": [
     "c",
     1.897
    ],
    "easy-intent-11": [
     "c",
     0.406
    ],
    "easy-fact-00": [
     "c",
     2.128
    ],
    "easy-fact-01": [
     "c",
     1.421
    ],
    "easy-fact-02": [
     "c",
     2.135
    ],
    "easy-fact-03": [
     "c",
     1.892
    ],
    "easy-fact-04": [
     "c",
     2.124
    ],
    "easy-fact-05": [
     "c",
     1.214
    ],
    "easy-fact-06": [
     "c",
     1.927
    ],
    "easy-fact-07": [
     "c",
     2.131
    ],
    "easy-fact-08": [
     "c",
     1.412
    ],
    "easy-fact-09": [
     "c",
     0.408
    ],
    "easy-fact-10": [
     "c",
     1.212
    ],
    "easy-fact-11": [
     "c",
     2.119
    ],
    "easy-extraction-00": [
     "c",
     1.209
    ],
    "easy-extraction-01": [
     "c",
     1.21
    ],
    "easy-extraction-02": [
     "c",
     1.419
    ],
    "easy-extraction-03": [
     "c",
     1.928
    ],
    "easy-extraction-04": [
     "c",
     1.211
    ],
    "easy-extraction-05": [
     "c",
     0.407
    ],
    "easy-extraction-06": [
     "c",
     1.436
    ],
    "easy-extraction-07": [
     "c",
     1.4
    ],
    "easy-extraction-08": [
     "c",
     1.209
    ],
    "easy-extraction-09": [
     "c",
     1.217
    ],
    "easy-extraction-10": [
     "c",
     1.178
    ],
    "easy-extraction-11": [
     "c",
     1.202
    ],
    "easy-tool_selection-00": [
     "c",
     1.414
    ],
    "easy-tool_selection-01": [
     "c",
     1.196
    ],
    "easy-tool_selection-02": [
     "c",
     1.405
    ],
    "easy-tool_selection-03": [
     "c",
     1.411
    ],
    "easy-tool_selection-04": [
     "c",
     1.405
    ],
    "easy-tool_selection-05": [
     "c",
     0.409
    ],
    "easy-tool_selection-06": [
     "c",
     1.425
    ],
    "easy-tool_selection-07": [
     "c",
     0.407
    ],
    "easy-tool_selection-08": [
     "c",
     1.203
    ],
    "easy-tool_selection-09": [
     "c",
     1.408
    ],
    "easy-tool_selection-10": [
     "c",
     0.406
    ],
    "easy-tool_selection-11": [
     "c",
     0.407
    ],
    "original-policy-01-0": [
     "c",
     2.178
    ],
    "original-policy-01-1": [
     "c",
     1.927
    ],
    "original-policy-02-0": [
     "c",
     1.204
    ],
    "original-policy-02-1": [
     "w",
     1.429
    ],
    "original-policy-03-0": [
     "c",
     1.922
    ],
    "original-policy-03-1": [
     "c",
     1.413
    ],
    "original-policy-04-0": [
     "c",
     1.907
    ],
    "original-policy-04-1": [
     "c",
     2.101
    ],
    "original-policy-05-0": [
     "c",
     1.887
    ],
    "original-policy-05-1": [
     "c",
     1.189
    ],
    "original-policy-06-0": [
     "c",
     0.408
    ],
    "original-policy-06-1": [
     "w",
     0.408
    ],
    "original-intent-01-0": [
     "c",
     1.202
    ],
    "original-intent-01-1": [
     "c",
     1.402
    ],
    "original-intent-02-0": [
     "c",
     0.408
    ],
    "original-intent-02-1": [
     "c",
     0.409
    ],
    "original-intent-03-0": [
     "c",
     1.203
    ],
    "original-intent-03-1": [
     "c",
     0.409
    ],
    "original-intent-04-0": [
     "c",
     1.428
    ],
    "original-intent-04-1": [
     "c",
     0.409
    ],
    "original-intent-05-0": [
     "c",
     0.41
    ],
    "original-intent-05-1": [
     "c",
     1.417
    ],
    "original-intent-06-0": [
     "c",
     1.209
    ],
    "original-intent-06-1": [
     "c",
     1.42
    ],
    "original-ordinal-01-0": [
     "c",
     1.449
    ],
    "original-ordinal-01-1": [
     "c",
     1.21
    ],
    "original-ordinal-02-0": [
     "c",
     1.478
    ],
    "original-ordinal-02-1": [
     "c",
     1.21
    ],
    "original-ordinal-03-0": [
     "c",
     0.409
    ],
    "original-ordinal-03-1": [
     "c",
     1.205
    ],
    "original-ordinal-04-0": [
     "c",
     0.694
    ],
    "original-ordinal-04-1": [
     "c",
     1.487
    ],
    "original-ordinal-05-0": [
     "c",
     1.492
    ],
    "original-ordinal-05-1": [
     "c",
     0.414
    ],
    "original-ordinal-06-0": [
     "c",
     0.415
    ],
    "original-ordinal-06-1": [
     "c",
     2.027
    ],
    "original-extraction-01-0": [
     "c",
     1.496
    ],
    "original-extraction-01-1": [
     "c",
     1.481
    ],
    "original-extraction-02-0": [
     "c",
     1.25
    ],
    "original-extraction-02-1": [
     "c",
     0.411
    ],
    "original-extraction-03-0": [
     "c",
     1.459
    ],
    "original-extraction-03-1": [
     "c",
     1.239
    ],
    "original-extraction-04-0": [
     "c",
     0.408
    ],
    "original-extraction-04-1": [
     "c",
     0.41
    ],
    "original-extraction-05-0": [
     "c",
     0.408
    ],
    "original-extraction-05-1": [
     "c",
     1.241
    ],
    "original-extraction-06-0": [
     "c",
     0.409
    ],
    "original-extraction-06-1": [
     "c",
     0.408
    ],
    "original-adequacy-01-0": [
     "c",
     1.454
    ],
    "original-adequacy-01-1": [
     "c",
     1.451
    ],
    "original-adequacy-02-0": [
     "c",
     1.232
    ],
    "original-adequacy-02-1": [
     "c",
     1.459
    ],
    "original-adequacy-03-0": [
     "c",
     0.409
    ],
    "original-adequacy-03-1": [
     "w",
     1.452
    ],
    "original-adequacy-04-0": [
     "c",
     0.409
    ],
    "original-adequacy-04-1": [
     "c",
     0.409
    ],
    "original-adequacy-05-0": [
     "c",
     0.41
    ],
    "original-adequacy-05-1": [
     "c",
     1.48
    ],
    "original-adequacy-06-0": [
     "c",
     1.233
    ],
    "original-adequacy-06-1": [
     "c",
     1.969
    ],
    "original-routing-01-0": [
     "c",
     1.244
    ],
    "original-routing-01-1": [
     "c",
     1.229
    ],
    "original-routing-02-0": [
     "c",
     1.442
    ],
    "original-routing-02-1": [
     "c",
     1.444
    ],
    "original-routing-03-0": [
     "c",
     0.41
    ],
    "original-routing-03-1": [
     "c",
     1.228
    ],
    "original-routing-04-0": [
     "c",
     0.41
    ],
    "original-routing-04-1": [
     "c",
     0.41
    ],
    "original-routing-05-0": [
     "c",
     1.446
    ],
    "original-routing-05-1": [
     "c",
     0.41
    ],
    "original-routing-06-0": [
     "c",
     0.408
    ],
    "original-routing-06-1": [
     "c",
     0.409
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     2.866
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     2.513
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     2.453
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     2.513
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     2.464
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     2.496
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     2.246
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     2.252
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.943
    ],
    "hard-opus-a-probability-04": [
     "w",
     2.045
    ],
    "hard-opus-a-probability-07": [
     "c",
     2.051
    ],
    "hard-opus-a-probability-08": [
     "c",
     1.938
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     1.954
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     2.181
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     2.248
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     1.917
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     2.142
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     1.917
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     1.906
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     2.149
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     1.464
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     1.936
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     1.927
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     1.957
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     2.572
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     2.448
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     2.283
    ],
    "hard-opus-b-multi_hop-07": [
     "w",
     2.299
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     2.458
    ],
    "hard-opus-b-probability-01": [
     "c",
     2.395
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.89
    ],
    "hard-opus-b-probability-03": [
     "c",
     2.036
    ],
    "hard-opus-b-probability-04": [
     "c",
     2.157
    ],
    "hard-opus-b-probability-06": [
     "c",
     2.201
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     1.95
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     1.921
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     2.149
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     1.942
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     1.943
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     1.462
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     2.788
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     2.563
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     2.743
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     2.744
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     2.72
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     2.785
    ],
    "hard-opus-c-long_policy-11": [
     "c",
     2.532
    ],
    "hard-opus-c-probability-03": [
     "c",
     1.936
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     2.218
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     2.214
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.984
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     2.216
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     1.974
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     1.945
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     1.19
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     1.199
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.188
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     1.422
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     2.098
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     2.116
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     2.302
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     2.277
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     2.3
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     2.508
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     2.521
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     2.491
    ],
    "hard-sol-a-multi_hop-10": [
     "c",
     2.509
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     2.283
    ],
    "hard-sol-a-trap-02": [
     "c",
     2.121
    ],
    "hard-sol-a-trap-04": [
     "c",
     2.132
    ],
    "hard-sol-a-trap-05": [
     "c",
     1.923
    ],
    "hard-sol-a-trap-06": [
     "c",
     1.191
    ],
    "hard-sol-a-trap-08": [
     "c",
     2.112
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.904
    ],
    "hard-sol-a-trap-13": [
     "c",
     2.089
    ],
    "hard-sol-a-trap-15": [
     "c",
     2.117
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     2.089
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     2.101
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     1.404
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     2.109
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     1.876
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     2.109
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     2.114
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     2.113
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     2.416
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     2.625
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     2.399
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     2.634
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.949
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     2.122
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     1.401
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     1.923
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.884
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.412
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     1.21
    ],
    "hard-sol-b-temporal_numeric-03": [
     "c",
     2.079
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     1.4
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     1.89
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     1.208
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     1.895
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     2.089
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     2.115
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     1.869
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.408
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     2.125
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.176
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     2.474
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     2.221
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     2.229
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     2.098
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     2.437
    ]
   }
  },
  "reflex-4b": {
   "display": "reflex 4B (kshetrajna12)",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 91,
     "w": 5,
     "f": 0,
     "n": 0,
     "accuracy": 0.9479
    },
    "judge": {
     "c": 142,
     "w": 4,
     "f": 0,
     "n": 0,
     "accuracy": 0.9726
    },
    "hard": {
     "c": 139,
     "w": 81,
     "f": 0,
     "n": 0,
     "accuracy": 0.6318
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     }
    },
    "judge": {
     "routing": {
      "c": 77,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9872
     },
     "adequacy": {
      "c": 65,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.9559
     }
    },
    "hard": {
     "long_policy": {
      "c": 20,
      "w": 18,
      "f": 0,
      "n": 0,
      "accuracy": 0.5263
     },
     "probability": {
      "c": 11,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.55
     },
     "temporal_numeric": {
      "c": 10,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "ambiguous": {
      "c": 7,
      "w": 7,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "multi_hop": {
      "c": 22,
      "w": 13,
      "f": 0,
      "n": 0,
      "accuracy": 0.6286
     },
     "tradeoff": {
      "c": 9,
      "w": 3,
      "f": 0,
      "n": 0,
      "accuracy": 0.75
     },
     "adversarial": {
      "c": 10,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8333
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 24,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.7273
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.331
    ],
    "easy-intent-01": [
     "c",
     0.33
    ],
    "easy-intent-02": [
     "c",
     0.33
    ],
    "easy-intent-03": [
     "c",
     0.33
    ],
    "easy-intent-04": [
     "c",
     0.329
    ],
    "easy-intent-05": [
     "c",
     0.33
    ],
    "easy-intent-06": [
     "c",
     0.33
    ],
    "easy-intent-07": [
     "c",
     0.33
    ],
    "easy-intent-08": [
     "c",
     0.333
    ],
    "easy-intent-09": [
     "c",
     0.33
    ],
    "easy-intent-10": [
     "c",
     0.331
    ],
    "easy-intent-11": [
     "c",
     0.329
    ],
    "easy-fact-00": [
     "c",
     0.33
    ],
    "easy-fact-01": [
     "c",
     0.33
    ],
    "easy-fact-02": [
     "c",
     0.331
    ],
    "easy-fact-03": [
     "c",
     0.331
    ],
    "easy-fact-04": [
     "c",
     0.33
    ],
    "easy-fact-05": [
     "c",
     0.33
    ],
    "easy-fact-06": [
     "c",
     0.331
    ],
    "easy-fact-07": [
     "c",
     0.33
    ],
    "easy-fact-08": [
     "c",
     0.33
    ],
    "easy-fact-09": [
     "c",
     0.33
    ],
    "easy-fact-10": [
     "c",
     0.33
    ],
    "easy-fact-11": [
     "c",
     0.331
    ],
    "easy-extraction-00": [
     "c",
     0.33
    ],
    "easy-extraction-01": [
     "c",
     0.332
    ],
    "easy-extraction-02": [
     "c",
     0.33
    ],
    "easy-extraction-03": [
     "c",
     0.33
    ],
    "easy-extraction-04": [
     "c",
     0.331
    ],
    "easy-extraction-05": [
     "c",
     0.329
    ],
    "easy-extraction-06": [
     "c",
     0.33
    ],
    "easy-extraction-07": [
     "c",
     0.331
    ],
    "easy-extraction-08": [
     "c",
     0.332
    ],
    "easy-extraction-09": [
     "c",
     0.331
    ],
    "easy-extraction-10": [
     "c",
     0.331
    ],
    "easy-extraction-11": [
     "c",
     0.33
    ],
    "easy-tool_selection-00": [
     "c",
     0.331
    ],
    "easy-tool_selection-01": [
     "c",
     0.332
    ],
    "easy-tool_selection-02": [
     "c",
     0.329
    ],
    "easy-tool_selection-03": [
     "c",
     0.338
    ],
    "easy-tool_selection-04": [
     "c",
     0.33
    ],
    "easy-tool_selection-05": [
     "c",
     0.328
    ],
    "easy-tool_selection-06": [
     "c",
     0.33
    ],
    "easy-tool_selection-07": [
     "c",
     0.331
    ],
    "easy-tool_selection-08": [
     "c",
     0.33
    ],
    "easy-tool_selection-09": [
     "c",
     0.329
    ],
    "easy-tool_selection-10": [
     "c",
     0.33
    ],
    "easy-tool_selection-11": [
     "c",
     0.33
    ],
    "original-policy-01-0": [
     "c",
     0.331
    ],
    "original-policy-01-1": [
     "c",
     2.076
    ],
    "original-policy-02-0": [
     "c",
     1.325
    ],
    "original-policy-02-1": [
     "w",
     1.112
    ],
    "original-policy-03-0": [
     "c",
     2.018
    ],
    "original-policy-03-1": [
     "c",
     1.105
    ],
    "original-policy-04-0": [
     "c",
     2.006
    ],
    "original-policy-04-1": [
     "c",
     1.794
    ],
    "original-policy-05-0": [
     "c",
     2.018
    ],
    "original-policy-05-1": [
     "c",
     1.332
    ],
    "original-policy-06-0": [
     "c",
     0.33
    ],
    "original-policy-06-1": [
     "c",
     0.329
    ],
    "original-intent-01-0": [
     "c",
     1.139
    ],
    "original-intent-01-1": [
     "c",
     1.334
    ],
    "original-intent-02-0": [
     "c",
     0.329
    ],
    "original-intent-02-1": [
     "c",
     0.33
    ],
    "original-intent-03-0": [
     "c",
     1.136
    ],
    "original-intent-03-1": [
     "c",
     0.331
    ],
    "original-intent-04-0": [
     "c",
     1.345
    ],
    "original-intent-04-1": [
     "c",
     0.331
    ],
    "original-intent-05-0": [
     "c",
     0.33
    ],
    "original-intent-05-1": [
     "c",
     1.347
    ],
    "original-intent-06-0": [
     "c",
     1.14
    ],
    "original-intent-06-1": [
     "c",
     1.343
    ],
    "original-ordinal-01-0": [
     "c",
     1.367
    ],
    "original-ordinal-01-1": [
     "c",
     1.127
    ],
    "original-ordinal-02-0": [
     "c",
     1.34
    ],
    "original-ordinal-02-1": [
     "c",
     1.126
    ],
    "original-ordinal-03-0": [
     "c",
     0.331
    ],
    "original-ordinal-03-1": [
     "c",
     1.142
    ],
    "original-ordinal-04-0": [
     "c",
     0.331
    ],
    "original-ordinal-04-1": [
     "c",
     1.343
    ],
    "original-ordinal-05-0": [
     "c",
     1.351
    ],
    "original-ordinal-05-1": [
     "c",
     0.331
    ],
    "original-ordinal-06-0": [
     "c",
     0.331
    ],
    "original-ordinal-06-1": [
     "c",
     1.84
    ],
    "original-extraction-01-0": [
     "c",
     1.354
    ],
    "original-extraction-01-1": [
     "c",
     1.339
    ],
    "original-extraction-02-0": [
     "c",
     1.13
    ],
    "original-extraction-02-1": [
     "c",
     0.329
    ],
    "original-extraction-03-0": [
     "c",
     1.335
    ],
    "original-extraction-03-1": [
     "c",
     1.132
    ],
    "original-extraction-04-0": [
     "c",
     0.329
    ],
    "original-extraction-04-1": [
     "c",
     0.33
    ],
    "original-extraction-05-0": [
     "c",
     0.329
    ],
    "original-extraction-05-1": [
     "c",
     1.143
    ],
    "original-extraction-06-0": [
     "c",
     0.331
    ],
    "original-extraction-06-1": [
     "c",
     0.33
    ],
    "original-adequacy-01-0": [
     "c",
     1.139
    ],
    "original-adequacy-01-1": [
     "c",
     1.106
    ],
    "original-adequacy-02-0": [
     "c",
     1.319
    ],
    "original-adequacy-02-1": [
     "c",
     1.101
    ],
    "original-adequacy-03-0": [
     "w",
     0.33
    ],
    "original-adequacy-03-1": [
     "c",
     1.104
    ],
    "original-adequacy-04-0": [
     "c",
     0.331
    ],
    "original-adequacy-04-1": [
     "c",
     0.336
    ],
    "original-adequacy-05-0": [
     "c",
     1.341
    ],
    "original-adequacy-05-1": [
     "w",
     1.104
    ],
    "original-adequacy-06-0": [
     "c",
     1.333
    ],
    "original-adequacy-06-1": [
     "c",
     1.996
    ],
    "original-routing-01-0": [
     "c",
     1.134
    ],
    "original-routing-01-1": [
     "c",
     1.104
    ],
    "original-routing-02-0": [
     "c",
     1.325
    ],
    "original-routing-02-1": [
     "c",
     1.339
    ],
    "original-routing-03-0": [
     "c",
     0.329
    ],
    "original-routing-03-1": [
     "c",
     1.131
    ],
    "original-routing-04-0": [
     "c",
     0.329
    ],
    "original-routing-04-1": [
     "w",
     0.33
    ],
    "original-routing-05-0": [
     "c",
     1.339
    ],
    "original-routing-05-1": [
     "c",
     0.331
    ],
    "original-routing-06-0": [
     "c",
     0.33
    ],
    "original-routing-06-1": [
     "c",
     0.33
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     2.29
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     2.072
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     2.046
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     2.068
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.833
    ],
    "hard-opus-a-long_policy-13": [
     "c",
     1.871
    ],
    "hard-opus-a-long_policy-17": [
     "w",
     1.833
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     2.054
    ],
    "hard-opus-a-probability-03": [
     "w",
     2.032
    ],
    "hard-opus-a-probability-04": [
     "w",
     1.833
    ],
    "hard-opus-a-probability-07": [
     "w",
     1.826
    ],
    "hard-opus-a-probability-08": [
     "c",
     2.094
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     2.039
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     1.827
    ],
    "hard-opus-a-temporal_numeric-07": [
     "c",
     2.062
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     2.013
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     1.994
    ],
    "hard-opus-b-ambiguous-02": [
     "w",
     1.79
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     1.798
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     2.008
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     1.331
    ],
    "hard-opus-b-ambiguous-10": [
     "w",
     1.798
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     1.826
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     1.817
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     2.087
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     2.085
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     1.877
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     1.879
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     1.906
    ],
    "hard-opus-b-probability-01": [
     "c",
     2.069
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.781
    ],
    "hard-opus-b-probability-03": [
     "w",
     1.996
    ],
    "hard-opus-b-probability-04": [
     "c",
     1.782
    ],
    "hard-opus-b-probability-06": [
     "c",
     1.999
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     1.787
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     1.804
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     1.789
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     1.799
    ],
    "hard-opus-b-tradeoff-08": [
     "c",
     2.008
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     1.337
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     2.218
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     2.009
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     2.006
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     2.225
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     2.255
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     2.242
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     2.236
    ],
    "hard-opus-c-probability-03": [
     "c",
     1.829
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     1.851
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     2.067
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.837
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     2.04
    ],
    "hard-opus-c-temporal_numeric-08": [
     "c",
     1.841
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     1.839
    ],
    "hard-sol-a-adversarial-01": [
     "w",
     1.141
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     1.342
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.135
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     1.127
    ],
    "hard-sol-a-adversarial-09": [
     "w",
     2.068
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     2.063
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     1.9
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     1.881
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     1.884
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     2.073
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     2.088
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     2.079
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     2.06
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     1.871
    ],
    "hard-sol-a-trap-02": [
     "c",
     1.799
    ],
    "hard-sol-a-trap-04": [
     "c",
     1.8
    ],
    "hard-sol-a-trap-05": [
     "c",
     1.832
    ],
    "hard-sol-a-trap-06": [
     "c",
     1.125
    ],
    "hard-sol-a-trap-08": [
     "c",
     2.04
    ],
    "hard-sol-a-trap-10": [
     "c",
     1.829
    ],
    "hard-sol-a-trap-13": [
     "c",
     1.835
    ],
    "hard-sol-a-trap-15": [
     "c",
     1.839
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.831
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     1.835
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     1.14
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     1.881
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     2.06
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     1.8
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     1.799
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     1.776
    ],
    "hard-sol-b-long_policy-01": [
     "w",
     1.958
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     2.189
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     1.963
    ],
    "hard-sol-b-long_policy-06": [
     "w",
     2.213
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.82
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     2.044
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     1.309
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     1.785
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.798
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.312
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     1.12
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     2.006
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     1.33
    ],
    "hard-sol-c-judge_hard-03": [
     "w",
     2.056
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     1.328
    ],
    "hard-sol-c-judge_hard-07": [
     "c",
     2.005
    ],
    "hard-sol-c-judge_hard-08": [
     "w",
     2.088
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     1.863
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     2.081
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.332
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     1.854
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.355
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     2.11
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     1.894
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     1.904
    ],
    "hard-sol-c-multi_hop-12": [
     "w",
     2.053
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     2.1
    ]
   }
  },
  "simplejev-qwen3.6-35b-a3b": {
   "display": "SimpleJev Qwen3.6-35B-A3B",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 90,
     "w": 6,
     "f": 0,
     "n": 0,
     "accuracy": 0.9375
    },
    "judge": {
     "c": 136,
     "w": 10,
     "f": 0,
     "n": 0,
     "accuracy": 0.9315
    },
    "hard": {
     "c": 146,
     "w": 74,
     "f": 0,
     "n": 0,
     "accuracy": 0.6636
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 11,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.6875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 72,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.9231
     },
     "adequacy": {
      "c": 64,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.9412
     }
    },
    "hard": {
     "long_policy": {
      "c": 23,
      "w": 15,
      "f": 0,
      "n": 0,
      "accuracy": 0.6053
     },
     "probability": {
      "c": 10,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.5
     },
     "temporal_numeric": {
      "c": 10,
      "w": 20,
      "f": 0,
      "n": 0,
      "accuracy": 0.3333
     },
     "ambiguous": {
      "c": 8,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.5714
     },
     "multi_hop": {
      "c": 26,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.7429
     },
     "tradeoff": {
      "c": 8,
      "w": 4,
      "f": 0,
      "n": 0,
      "accuracy": 0.6667
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "judge_hard": {
      "c": 24,
      "w": 9,
      "f": 0,
      "n": 0,
      "accuracy": 0.7273
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     0.929
    ],
    "easy-intent-01": [
     "c",
     0.994
    ],
    "easy-intent-02": [
     "c",
     0.903
    ],
    "easy-intent-03": [
     "c",
     0.883
    ],
    "easy-intent-04": [
     "c",
     0.905
    ],
    "easy-intent-05": [
     "c",
     0.844
    ],
    "easy-intent-06": [
     "c",
     0.556
    ],
    "easy-intent-07": [
     "c",
     0.93
    ],
    "easy-intent-08": [
     "c",
     0.914
    ],
    "easy-intent-09": [
     "c",
     0.854
    ],
    "easy-intent-10": [
     "c",
     0.948
    ],
    "easy-intent-11": [
     "c",
     0.887
    ],
    "easy-fact-00": [
     "c",
     0.867
    ],
    "easy-fact-01": [
     "c",
     0.869
    ],
    "easy-fact-02": [
     "c",
     0.939
    ],
    "easy-fact-03": [
     "c",
     0.52
    ],
    "easy-fact-04": [
     "c",
     0.872
    ],
    "easy-fact-05": [
     "c",
     0.846
    ],
    "easy-fact-06": [
     "c",
     0.551
    ],
    "easy-fact-07": [
     "c",
     0.947
    ],
    "easy-fact-08": [
     "c",
     0.839
    ],
    "easy-fact-09": [
     "c",
     0.515
    ],
    "easy-fact-10": [
     "c",
     0.898
    ],
    "easy-fact-11": [
     "c",
     0.927
    ],
    "easy-extraction-00": [
     "c",
     0.847
    ],
    "easy-extraction-01": [
     "c",
     0.953
    ],
    "easy-extraction-02": [
     "c",
     0.894
    ],
    "easy-extraction-03": [
     "c",
     0.571
    ],
    "easy-extraction-04": [
     "c",
     0.872
    ],
    "easy-extraction-05": [
     "c",
     0.886
    ],
    "easy-extraction-06": [
     "c",
     0.546
    ],
    "easy-extraction-07": [
     "c",
     0.883
    ],
    "easy-extraction-08": [
     "c",
     0.838
    ],
    "easy-extraction-09": [
     "c",
     0.856
    ],
    "easy-extraction-10": [
     "c",
     0.567
    ],
    "easy-extraction-11": [
     "c",
     0.91
    ],
    "easy-tool_selection-00": [
     "c",
     0.826
    ],
    "easy-tool_selection-01": [
     "c",
     0.829
    ],
    "easy-tool_selection-02": [
     "c",
     0.543
    ],
    "easy-tool_selection-03": [
     "c",
     0.544
    ],
    "easy-tool_selection-04": [
     "c",
     0.894
    ],
    "easy-tool_selection-05": [
     "c",
     0.859
    ],
    "easy-tool_selection-06": [
     "c",
     0.544
    ],
    "easy-tool_selection-07": [
     "c",
     0.515
    ],
    "easy-tool_selection-08": [
     "c",
     0.907
    ],
    "easy-tool_selection-09": [
     "c",
     0.914
    ],
    "easy-tool_selection-10": [
     "c",
     0.928
    ],
    "easy-tool_selection-11": [
     "c",
     0.866
    ],
    "original-policy-01-0": [
     "c",
     0.835
    ],
    "original-policy-01-1": [
     "c",
     0.568
    ],
    "original-policy-02-0": [
     "c",
     0.878
    ],
    "original-policy-02-1": [
     "c",
     0.869
    ],
    "original-policy-03-0": [
     "c",
     0.862
    ],
    "original-policy-03-1": [
     "c",
     0.687
    ],
    "original-policy-04-0": [
     "c",
     0.556
    ],
    "original-policy-04-1": [
     "c",
     0.903
    ],
    "original-policy-05-0": [
     "c",
     0.542
    ],
    "original-policy-05-1": [
     "c",
     0.882
    ],
    "original-policy-06-0": [
     "c",
     0.849
    ],
    "original-policy-06-1": [
     "w",
     0.845
    ],
    "original-intent-01-0": [
     "c",
     0.867
    ],
    "original-intent-01-1": [
     "c",
     0.867
    ],
    "original-intent-02-0": [
     "c",
     0.845
    ],
    "original-intent-02-1": [
     "c",
     0.851
    ],
    "original-intent-03-0": [
     "c",
     0.883
    ],
    "original-intent-03-1": [
     "c",
     0.834
    ],
    "original-intent-04-0": [
     "c",
     0.869
    ],
    "original-intent-04-1": [
     "c",
     0.519
    ],
    "original-intent-05-0": [
     "c",
     0.886
    ],
    "original-intent-05-1": [
     "c",
     0.857
    ],
    "original-intent-06-0": [
     "c",
     0.539
    ],
    "original-intent-06-1": [
     "c",
     0.839
    ],
    "original-ordinal-01-0": [
     "c",
     0.883
    ],
    "original-ordinal-01-1": [
     "c",
     0.532
    ],
    "original-ordinal-02-0": [
     "c",
     0.545
    ],
    "original-ordinal-02-1": [
     "c",
     0.837
    ],
    "original-ordinal-03-0": [
     "c",
     0.54
    ],
    "original-ordinal-03-1": [
     "c",
     0.851
    ],
    "original-ordinal-04-0": [
     "c",
     0.512
    ],
    "original-ordinal-04-1": [
     "c",
     0.914
    ],
    "original-ordinal-05-0": [
     "c",
     0.505
    ],
    "original-ordinal-05-1": [
     "c",
     0.838
    ],
    "original-ordinal-06-0": [
     "c",
     0.527
    ],
    "original-ordinal-06-1": [
     "c",
     0.864
    ],
    "original-extraction-01-0": [
     "c",
     0.867
    ],
    "original-extraction-01-1": [
     "c",
     0.563
    ],
    "original-extraction-02-0": [
     "c",
     0.899
    ],
    "original-extraction-02-1": [
     "c",
     0.509
    ],
    "original-extraction-03-0": [
     "c",
     0.843
    ],
    "original-extraction-03-1": [
     "c",
     0.534
    ],
    "original-extraction-04-0": [
     "c",
     0.855
    ],
    "original-extraction-04-1": [
     "c",
     0.498
    ],
    "original-extraction-05-0": [
     "c",
     0.889
    ],
    "original-extraction-05-1": [
     "c",
     0.871
    ],
    "original-extraction-06-0": [
     "c",
     0.512
    ],
    "original-extraction-06-1": [
     "c",
     0.557
    ],
    "original-adequacy-01-0": [
     "c",
     0.894
    ],
    "original-adequacy-01-1": [
     "c",
     0.844
    ],
    "original-adequacy-02-0": [
     "c",
     0.509
    ],
    "original-adequacy-02-1": [
     "c",
     0.828
    ],
    "original-adequacy-03-0": [
     "w",
     0.853
    ],
    "original-adequacy-03-1": [
     "w",
     0.554
    ],
    "original-adequacy-04-0": [
     "c",
     0.845
    ],
    "original-adequacy-04-1": [
     "c",
     0.892
    ],
    "original-adequacy-05-0": [
     "w",
     0.922
    ],
    "original-adequacy-05-1": [
     "c",
     0.908
    ],
    "original-adequacy-06-0": [
     "c",
     0.91
    ],
    "original-adequacy-06-1": [
     "w",
     0.525
    ],
    "original-routing-01-0": [
     "c",
     0.885
    ],
    "original-routing-01-1": [
     "c",
     0.553
    ],
    "original-routing-02-0": [
     "c",
     0.574
    ],
    "original-routing-02-1": [
     "c",
     0.891
    ],
    "original-routing-03-0": [
     "c",
     0.549
    ],
    "original-routing-03-1": [
     "c",
     0.837
    ],
    "original-routing-04-0": [
     "c",
     0.859
    ],
    "original-routing-04-1": [
     "c",
     0.528
    ],
    "original-routing-05-0": [
     "c",
     0.54
    ],
    "original-routing-05-1": [
     "c",
     0.879
    ],
    "original-routing-06-0": [
     "c",
     0.931
    ],
    "original-routing-06-1": [
     "c",
     0.571
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     0.734
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     0.641
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     0.927
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     0.947
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     0.935
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     0.647
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     0.897
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     0.598
    ],
    "hard-opus-a-probability-03": [
     "c",
     0.545
    ],
    "hard-opus-a-probability-04": [
     "w",
     0.92
    ],
    "hard-opus-a-probability-07": [
     "c",
     0.849
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.897
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     0.86
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     0.888
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.886
    ],
    "hard-opus-a-temporal_numeric-09": [
     "w",
     0.871
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     0.901
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     0.522
    ],
    "hard-opus-b-ambiguous-03": [
     "w",
     0.534
    ],
    "hard-opus-b-ambiguous-07": [
     "w",
     0.504
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     0.571
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     0.843
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.9
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.83
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     0.954
    ],
    "hard-opus-b-multi_hop-04": [
     "w",
     0.993
    ],
    "hard-opus-b-multi_hop-05": [
     "c",
     0.585
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.949
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     0.627
    ],
    "hard-opus-b-probability-01": [
     "w",
     0.975
    ],
    "hard-opus-b-probability-02": [
     "w",
     0.576
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.949
    ],
    "hard-opus-b-probability-04": [
     "c",
     0.556
    ],
    "hard-opus-b-probability-06": [
     "w",
     0.913
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.572
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.831
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.564
    ],
    "hard-opus-b-tradeoff-07": [
     "w",
     0.888
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     0.893
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.528
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     1.012
    ],
    "hard-opus-c-long_policy-03": [
     "w",
     0.602
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     0.92
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     0.926
    ],
    "hard-opus-c-long_policy-08": [
     "c",
     0.663
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     0.964
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     0.997
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.88
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     0.894
    ],
    "hard-opus-c-temporal_numeric-03": [
     "c",
     0.867
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     0.894
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     0.901
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     0.902
    ],
    "hard-opus-c-temporal_numeric-12": [
     "c",
     0.969
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     0.882
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     0.904
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     0.883
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     0.873
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     0.878
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     0.917
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     0.986
    ],
    "hard-sol-a-multi_hop-02": [
     "c",
     0.96
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     0.941
    ],
    "hard-sol-a-multi_hop-07": [
     "w",
     0.935
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     0.964
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     0.915
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.895
    ],
    "hard-sol-a-multi_hop-12": [
     "w",
     0.983
    ],
    "hard-sol-a-trap-02": [
     "c",
     0.621
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.826
    ],
    "hard-sol-a-trap-05": [
     "c",
     0.924
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.839
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.917
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.548
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.885
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.868
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     0.916
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.61
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.867
    ],
    "hard-sol-b-judge_hard-08": [
     "w",
     0.856
    ],
    "hard-sol-b-judge_hard-10": [
     "w",
     0.55
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     0.822
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     0.864
    ],
    "hard-sol-b-judge_hard-18": [
     "w",
     0.864
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     0.974
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     0.917
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     0.927
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     1.014
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     0.551
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     0.823
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     0.847
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     0.936
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     0.6
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     0.853
    ],
    "hard-sol-b-temporal_numeric-02": [
     "w",
     0.903
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     0.919
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     0.504
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.868
    ],
    "hard-sol-c-judge_hard-05": [
     "c",
     0.839
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     0.583
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     0.495
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     0.902
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     0.546
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     0.855
    ],
    "hard-sol-c-judge_hard-13": [
     "w",
     0.915
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     0.95
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     0.582
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     0.619
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     0.987
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     0.84
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     0.929
    ]
   }
  },
  "simplejev-qwen3.8-27b": {
   "display": "SimpleJev Qwen3.8-27B",
   "partial": false,
   "by_tier": {
    "easy": {
     "c": 72,
     "w": 0,
     "f": 0,
     "n": 0,
     "accuracy": 1.0
    },
    "standard": {
     "c": 93,
     "w": 3,
     "f": 0,
     "n": 0,
     "accuracy": 0.9688
    },
    "judge": {
     "c": 136,
     "w": 10,
     "f": 0,
     "n": 0,
     "accuracy": 0.9315
    },
    "hard": {
     "c": 165,
     "w": 55,
     "f": 0,
     "n": 0,
     "accuracy": 0.75
    }
   },
   "by_tier_topic": {
    "easy": {
     "intent": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "fact": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "tool_selection": {
      "c": 18,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "standard": {
     "policy": {
      "c": 15,
      "w": 1,
      "f": 0,
      "n": 0,
      "accuracy": 0.9375
     },
     "intent": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "ordinal": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "extraction": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 14,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.875
     },
     "routing": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    },
    "judge": {
     "routing": {
      "c": 78,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "adequacy": {
      "c": 58,
      "w": 10,
      "f": 0,
      "n": 0,
      "accuracy": 0.8529
     }
    },
    "hard": {
     "long_policy": {
      "c": 26,
      "w": 12,
      "f": 0,
      "n": 0,
      "accuracy": 0.6842
     },
     "probability": {
      "c": 14,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.7
     },
     "temporal_numeric": {
      "c": 8,
      "w": 22,
      "f": 0,
      "n": 0,
      "accuracy": 0.2667
     },
     "ambiguous": {
      "c": 12,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8571
     },
     "multi_hop": {
      "c": 29,
      "w": 6,
      "f": 0,
      "n": 0,
      "accuracy": 0.8286
     },
     "tradeoff": {
      "c": 10,
      "w": 2,
      "f": 0,
      "n": 0,
      "accuracy": 0.8333
     },
     "adversarial": {
      "c": 12,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "trap": {
      "c": 16,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     },
     "judge_hard": {
      "c": 28,
      "w": 5,
      "f": 0,
      "n": 0,
      "accuracy": 0.8485
     },
     "routing_hard": {
      "c": 10,
      "w": 0,
      "f": 0,
      "n": 0,
      "accuracy": 1.0
     }
    }
   },
   "public_tasks": {
    "easy-intent-00": [
     "c",
     1.59
    ],
    "easy-intent-01": [
     "c",
     1.151
    ],
    "easy-intent-02": [
     "c",
     1.822
    ],
    "easy-intent-03": [
     "c",
     0.957
    ],
    "easy-intent-04": [
     "c",
     1.191
    ],
    "easy-intent-05": [
     "c",
     1.737
    ],
    "easy-intent-06": [
     "c",
     0.925
    ],
    "easy-intent-07": [
     "c",
     2.31
    ],
    "easy-intent-08": [
     "c",
     0.879
    ],
    "easy-intent-09": [
     "c",
     1.79
    ],
    "easy-intent-10": [
     "c",
     0.905
    ],
    "easy-intent-11": [
     "c",
     0.999
    ],
    "easy-fact-00": [
     "c",
     0.834
    ],
    "easy-fact-01": [
     "c",
     1.762
    ],
    "easy-fact-02": [
     "c",
     1.531
    ],
    "easy-fact-03": [
     "c",
     0.89
    ],
    "easy-fact-04": [
     "c",
     0.94
    ],
    "easy-fact-05": [
     "c",
     0.871
    ],
    "easy-fact-06": [
     "c",
     0.865
    ],
    "easy-fact-07": [
     "c",
     0.926
    ],
    "easy-fact-08": [
     "c",
     0.911
    ],
    "easy-fact-09": [
     "c",
     1.092
    ],
    "easy-fact-10": [
     "c",
     0.843
    ],
    "easy-fact-11": [
     "c",
     0.918
    ],
    "easy-extraction-00": [
     "c",
     0.882
    ],
    "easy-extraction-01": [
     "c",
     0.852
    ],
    "easy-extraction-02": [
     "c",
     0.866
    ],
    "easy-extraction-03": [
     "c",
     0.873
    ],
    "easy-extraction-04": [
     "c",
     0.931
    ],
    "easy-extraction-05": [
     "c",
     0.897
    ],
    "easy-extraction-06": [
     "c",
     1.213
    ],
    "easy-extraction-07": [
     "c",
     0.862
    ],
    "easy-extraction-08": [
     "c",
     1.157
    ],
    "easy-extraction-09": [
     "c",
     0.878
    ],
    "easy-extraction-10": [
     "c",
     1.889
    ],
    "easy-extraction-11": [
     "c",
     1.658
    ],
    "easy-tool_selection-00": [
     "c",
     1.2
    ],
    "easy-tool_selection-01": [
     "c",
     0.556
    ],
    "easy-tool_selection-02": [
     "c",
     1.069
    ],
    "easy-tool_selection-03": [
     "c",
     0.906
    ],
    "easy-tool_selection-04": [
     "c",
     0.845
    ],
    "easy-tool_selection-05": [
     "c",
     0.545
    ],
    "easy-tool_selection-06": [
     "c",
     1.75
    ],
    "easy-tool_selection-07": [
     "c",
     0.858
    ],
    "easy-tool_selection-08": [
     "c",
     2.196
    ],
    "easy-tool_selection-09": [
     "c",
     1.416
    ],
    "easy-tool_selection-10": [
     "c",
     1.663
    ],
    "easy-tool_selection-11": [
     "c",
     0.944
    ],
    "original-policy-01-0": [
     "c",
     0.9
    ],
    "original-policy-01-1": [
     "c",
     0.888
    ],
    "original-policy-02-0": [
     "c",
     0.85
    ],
    "original-policy-02-1": [
     "c",
     0.85
    ],
    "original-policy-03-0": [
     "c",
     0.868
    ],
    "original-policy-03-1": [
     "c",
     0.534
    ],
    "original-policy-04-0": [
     "c",
     0.862
    ],
    "original-policy-04-1": [
     "c",
     0.855
    ],
    "original-policy-05-0": [
     "c",
     0.876
    ],
    "original-policy-05-1": [
     "c",
     0.945
    ],
    "original-policy-06-0": [
     "c",
     1.312
    ],
    "original-policy-06-1": [
     "w",
     0.627
    ],
    "original-intent-01-0": [
     "c",
     0.908
    ],
    "original-intent-01-1": [
     "c",
     0.912
    ],
    "original-intent-02-0": [
     "c",
     0.53
    ],
    "original-intent-02-1": [
     "c",
     0.863
    ],
    "original-intent-03-0": [
     "c",
     0.858
    ],
    "original-intent-03-1": [
     "c",
     1.438
    ],
    "original-intent-04-0": [
     "c",
     0.858
    ],
    "original-intent-04-1": [
     "c",
     0.866
    ],
    "original-intent-05-0": [
     "c",
     1.324
    ],
    "original-intent-05-1": [
     "c",
     1.388
    ],
    "original-intent-06-0": [
     "c",
     0.834
    ],
    "original-intent-06-1": [
     "c",
     0.869
    ],
    "original-ordinal-01-0": [
     "c",
     1.241
    ],
    "original-ordinal-01-1": [
     "c",
     1.015
    ],
    "original-ordinal-02-0": [
     "c",
     1.61
    ],
    "original-ordinal-02-1": [
     "c",
     0.872
    ],
    "original-ordinal-03-0": [
     "c",
     1.152
    ],
    "original-ordinal-03-1": [
     "c",
     1.577
    ],
    "original-ordinal-04-0": [
     "c",
     0.894
    ],
    "original-ordinal-04-1": [
     "c",
     0.82
    ],
    "original-ordinal-05-0": [
     "c",
     1.817
    ],
    "original-ordinal-05-1": [
     "c",
     0.991
    ],
    "original-ordinal-06-0": [
     "c",
     1.531
    ],
    "original-ordinal-06-1": [
     "c",
     1.815
    ],
    "original-extraction-01-0": [
     "c",
     1.349
    ],
    "original-extraction-01-1": [
     "c",
     1.315
    ],
    "original-extraction-02-0": [
     "c",
     1.948
    ],
    "original-extraction-02-1": [
     "c",
     1.182
    ],
    "original-extraction-03-0": [
     "c",
     1.671
    ],
    "original-extraction-03-1": [
     "c",
     0.577
    ],
    "original-extraction-04-0": [
     "c",
     1.078
    ],
    "original-extraction-04-1": [
     "c",
     1.633
    ],
    "original-extraction-05-0": [
     "c",
     0.532
    ],
    "original-extraction-05-1": [
     "c",
     1.055
    ],
    "original-extraction-06-0": [
     "c",
     1.882
    ],
    "original-extraction-06-1": [
     "c",
     1.83
    ],
    "original-adequacy-01-0": [
     "c",
     0.922
    ],
    "original-adequacy-01-1": [
     "c",
     1.299
    ],
    "original-adequacy-02-0": [
     "c",
     1.648
    ],
    "original-adequacy-02-1": [
     "c",
     0.875
    ],
    "original-adequacy-03-0": [
     "c",
     0.531
    ],
    "original-adequacy-03-1": [
     "c",
     0.872
    ],
    "original-adequacy-04-0": [
     "c",
     0.916
    ],
    "original-adequacy-04-1": [
     "c",
     1.702
    ],
    "original-adequacy-05-0": [
     "c",
     0.935
    ],
    "original-adequacy-05-1": [
     "c",
     0.873
    ],
    "original-adequacy-06-0": [
     "c",
     1.84
    ],
    "original-adequacy-06-1": [
     "w",
     1.179
    ],
    "original-routing-01-0": [
     "c",
     1.314
    ],
    "original-routing-01-1": [
     "c",
     1.823
    ],
    "original-routing-02-0": [
     "c",
     1.393
    ],
    "original-routing-02-1": [
     "c",
     1.295
    ],
    "original-routing-03-0": [
     "c",
     0.879
    ],
    "original-routing-03-1": [
     "c",
     0.563
    ],
    "original-routing-04-0": [
     "c",
     0.847
    ],
    "original-routing-04-1": [
     "c",
     0.533
    ],
    "original-routing-05-0": [
     "c",
     0.855
    ],
    "original-routing-05-1": [
     "c",
     1.464
    ],
    "original-routing-06-0": [
     "c",
     1.078
    ],
    "original-routing-06-1": [
     "c",
     1.467
    ],
    "hard-opus-a-long_policy-01": [
     "c",
     2.078
    ],
    "hard-opus-a-long_policy-04": [
     "w",
     2.222
    ],
    "hard-opus-a-long_policy-08": [
     "c",
     1.052
    ],
    "hard-opus-a-long_policy-09": [
     "c",
     1.068
    ],
    "hard-opus-a-long_policy-11": [
     "c",
     1.795
    ],
    "hard-opus-a-long_policy-13": [
     "w",
     1.098
    ],
    "hard-opus-a-long_policy-17": [
     "c",
     2.341
    ],
    "hard-opus-a-long_policy-19": [
     "w",
     1.03
    ],
    "hard-opus-a-probability-03": [
     "c",
     1.261
    ],
    "hard-opus-a-probability-04": [
     "w",
     1.945
    ],
    "hard-opus-a-probability-07": [
     "c",
     2.254
    ],
    "hard-opus-a-probability-08": [
     "c",
     0.999
    ],
    "hard-opus-a-temporal_numeric-03": [
     "w",
     1.768
    ],
    "hard-opus-a-temporal_numeric-06": [
     "c",
     1.073
    ],
    "hard-opus-a-temporal_numeric-07": [
     "w",
     0.884
    ],
    "hard-opus-a-temporal_numeric-09": [
     "c",
     0.925
    ],
    "hard-opus-a-temporal_numeric-12": [
     "w",
     1.598
    ],
    "hard-opus-b-ambiguous-02": [
     "c",
     2.32
    ],
    "hard-opus-b-ambiguous-03": [
     "c",
     0.88
    ],
    "hard-opus-b-ambiguous-07": [
     "c",
     1.529
    ],
    "hard-opus-b-ambiguous-09": [
     "c",
     1.425
    ],
    "hard-opus-b-ambiguous-10": [
     "c",
     1.107
    ],
    "hard-opus-b-ambiguous-11": [
     "c",
     0.881
    ],
    "hard-opus-b-ambiguous-13": [
     "c",
     0.901
    ],
    "hard-opus-b-multi_hop-03": [
     "c",
     0.756
    ],
    "hard-opus-b-multi_hop-04": [
     "c",
     1.069
    ],
    "hard-opus-b-multi_hop-05": [
     "w",
     1.046
    ],
    "hard-opus-b-multi_hop-07": [
     "c",
     0.776
    ],
    "hard-opus-b-multi_hop-08": [
     "c",
     1.091
    ],
    "hard-opus-b-probability-01": [
     "c",
     1.028
    ],
    "hard-opus-b-probability-02": [
     "w",
     1.224
    ],
    "hard-opus-b-probability-03": [
     "c",
     0.657
    ],
    "hard-opus-b-probability-04": [
     "c",
     1.169
    ],
    "hard-opus-b-probability-06": [
     "c",
     0.567
    ],
    "hard-opus-b-tradeoff-01": [
     "w",
     0.954
    ],
    "hard-opus-b-tradeoff-03": [
     "c",
     0.935
    ],
    "hard-opus-b-tradeoff-06": [
     "c",
     0.902
    ],
    "hard-opus-b-tradeoff-07": [
     "c",
     0.732
    ],
    "hard-opus-b-tradeoff-08": [
     "w",
     2.29
    ],
    "hard-opus-b-tradeoff-12": [
     "c",
     0.934
    ],
    "hard-opus-c-long_policy-02": [
     "c",
     2.44
    ],
    "hard-opus-c-long_policy-03": [
     "c",
     1.19
    ],
    "hard-opus-c-long_policy-04": [
     "c",
     1.584
    ],
    "hard-opus-c-long_policy-05": [
     "w",
     1.138
    ],
    "hard-opus-c-long_policy-08": [
     "w",
     2.711
    ],
    "hard-opus-c-long_policy-10": [
     "c",
     1.507
    ],
    "hard-opus-c-long_policy-11": [
     "w",
     2.33
    ],
    "hard-opus-c-probability-03": [
     "c",
     0.936
    ],
    "hard-opus-c-temporal_numeric-02": [
     "w",
     2.246
    ],
    "hard-opus-c-temporal_numeric-03": [
     "w",
     0.87
    ],
    "hard-opus-c-temporal_numeric-04": [
     "w",
     1.838
    ],
    "hard-opus-c-temporal_numeric-06": [
     "w",
     1.609
    ],
    "hard-opus-c-temporal_numeric-08": [
     "w",
     1.665
    ],
    "hard-opus-c-temporal_numeric-12": [
     "w",
     1.431
    ],
    "hard-sol-a-adversarial-01": [
     "c",
     1.701
    ],
    "hard-sol-a-adversarial-06": [
     "c",
     1.861
    ],
    "hard-sol-a-adversarial-07": [
     "c",
     1.935
    ],
    "hard-sol-a-adversarial-08": [
     "c",
     1.554
    ],
    "hard-sol-a-adversarial-09": [
     "c",
     1.679
    ],
    "hard-sol-a-adversarial-11": [
     "c",
     1.311
    ],
    "hard-sol-a-multi_hop-01": [
     "c",
     1.084
    ],
    "hard-sol-a-multi_hop-02": [
     "w",
     1.076
    ],
    "hard-sol-a-multi_hop-05": [
     "c",
     1.059
    ],
    "hard-sol-a-multi_hop-07": [
     "c",
     2.038
    ],
    "hard-sol-a-multi_hop-08": [
     "w",
     1.226
    ],
    "hard-sol-a-multi_hop-09": [
     "c",
     1.964
    ],
    "hard-sol-a-multi_hop-10": [
     "w",
     0.993
    ],
    "hard-sol-a-multi_hop-12": [
     "c",
     1.244
    ],
    "hard-sol-a-trap-02": [
     "c",
     1.826
    ],
    "hard-sol-a-trap-04": [
     "c",
     0.851
    ],
    "hard-sol-a-trap-05": [
     "c",
     1.752
    ],
    "hard-sol-a-trap-06": [
     "c",
     0.557
    ],
    "hard-sol-a-trap-08": [
     "c",
     0.865
    ],
    "hard-sol-a-trap-10": [
     "c",
     0.604
    ],
    "hard-sol-a-trap-13": [
     "c",
     0.951
    ],
    "hard-sol-a-trap-15": [
     "c",
     0.902
    ],
    "hard-sol-b-judge_hard-01": [
     "c",
     1.08
    ],
    "hard-sol-b-judge_hard-02": [
     "w",
     0.932
    ],
    "hard-sol-b-judge_hard-05": [
     "c",
     0.883
    ],
    "hard-sol-b-judge_hard-08": [
     "c",
     0.857
    ],
    "hard-sol-b-judge_hard-10": [
     "c",
     0.536
    ],
    "hard-sol-b-judge_hard-14": [
     "c",
     1.857
    ],
    "hard-sol-b-judge_hard-15": [
     "c",
     1.691
    ],
    "hard-sol-b-judge_hard-18": [
     "c",
     1.821
    ],
    "hard-sol-b-long_policy-01": [
     "c",
     1.17
    ],
    "hard-sol-b-long_policy-02": [
     "c",
     1.384
    ],
    "hard-sol-b-long_policy-05": [
     "c",
     2.177
    ],
    "hard-sol-b-long_policy-06": [
     "c",
     2.238
    ],
    "hard-sol-b-routing_hard-01": [
     "c",
     1.845
    ],
    "hard-sol-b-routing_hard-02": [
     "c",
     1.348
    ],
    "hard-sol-b-routing_hard-03": [
     "c",
     2.129
    ],
    "hard-sol-b-routing_hard-07": [
     "c",
     2.165
    ],
    "hard-sol-b-routing_hard-09": [
     "c",
     1.302
    ],
    "hard-sol-b-temporal_numeric-01": [
     "w",
     1.743
    ],
    "hard-sol-b-temporal_numeric-02": [
     "c",
     1.432
    ],
    "hard-sol-b-temporal_numeric-03": [
     "w",
     2.063
    ],
    "hard-sol-b-temporal_numeric-04": [
     "w",
     1.754
    ],
    "hard-sol-c-judge_hard-03": [
     "c",
     0.965
    ],
    "hard-sol-c-judge_hard-05": [
     "w",
     2.033
    ],
    "hard-sol-c-judge_hard-07": [
     "w",
     1.937
    ],
    "hard-sol-c-judge_hard-08": [
     "c",
     1.196
    ],
    "hard-sol-c-judge_hard-09": [
     "c",
     1.413
    ],
    "hard-sol-c-judge_hard-10": [
     "c",
     1.171
    ],
    "hard-sol-c-judge_hard-11": [
     "c",
     1.127
    ],
    "hard-sol-c-judge_hard-13": [
     "c",
     1.089
    ],
    "hard-sol-c-judge_hard-15": [
     "c",
     1.592
    ],
    "hard-sol-c-multi_hop-07": [
     "c",
     1.759
    ],
    "hard-sol-c-multi_hop-09": [
     "c",
     1.269
    ],
    "hard-sol-c-multi_hop-11": [
     "c",
     2.297
    ],
    "hard-sol-c-multi_hop-12": [
     "c",
     1.153
    ],
    "hard-sol-c-multi_hop-13": [
     "c",
     1.287
    ]
   }
  }
 }
}
