[
  {
    "task": "feedback",
    "arm": "decisions",
    "method": "Posthoc strict loss < 0.50 minus 1e-10 numerical tolerance; original EM and same 2000 resamples",
    "point": {
      "naive": "qwen3:1.7b",
      "corr": "qwen3:0.6b",
      "oracle": "qwen3:0.6b"
    },
    "naiveMismatch": 0.3035,
    "correctedMismatch": 0.0445,
    "naive": 607,
    "corr": 89,
    "naiveBoundary": 323,
    "corrBoundary": 43,
    "oracleBoundary": 313,
    "missingClass": 0
  },
  {
    "task": "feedback",
    "arm": "responses",
    "method": "Posthoc strict loss < 0.50 minus 1e-10 numerical tolerance; original EM and same 2000 resamples",
    "point": {
      "naive": "qwen3:0.6b",
      "corr": "qwen3:0.6b",
      "oracle": "qwen3:0.6b"
    },
    "naiveMismatch": 0,
    "correctedMismatch": 0,
    "naive": 0,
    "corr": 0,
    "naiveBoundary": 313,
    "corrBoundary": 313,
    "oracleBoundary": 313,
    "missingClass": 0
  },
  {
    "task": "triage",
    "arm": "decisions",
    "method": "Posthoc strict loss < 0.50 minus 1e-10 numerical tolerance; original EM and same 2000 resamples",
    "point": {
      "naive": "qwen3:8b",
      "corr": "qwen3:8b",
      "oracle": "qwen3:8b"
    },
    "naiveMismatch": 0.29,
    "correctedMismatch": 0.237,
    "naive": 580,
    "corr": 474,
    "naiveBoundary": 492,
    "corrBoundary": 19,
    "oracleBoundary": 312,
    "missingClass": 0
  },
  {
    "task": "triage",
    "arm": "responses",
    "method": "Posthoc strict loss < 0.50 minus 1e-10 numerical tolerance; original EM and same 2000 resamples",
    "point": {
      "naive": "qwen3:8b",
      "corr": "qwen3:8b",
      "oracle": "qwen3:8b"
    },
    "naiveMismatch": 0.0055,
    "correctedMismatch": 0.1805,
    "naive": 11,
    "corr": 361,
    "naiveBoundary": 317,
    "corrBoundary": 110,
    "oracleBoundary": 312,
    "missingClass": 0
  }
]
