data/results.json

{
  "hero": "Production proof-advisor canary passed; Jensen theorem remained SEARCH_EXHAUSTED (0/6 Lean-valid). No RH proof.",
  "stages": [
    {
      "id": "stage1",
      "label": "SFT90",
      "status": "NO-GO baseline",
      "exact": null,
      "repo": "https://huggingface.co/FluffyAIcode/Kakeya-OProver-Stage1-SFT90"
    },
    {
      "id": "stage2",
      "label": "SFT1000",
      "status": "Matched baseline",
      "exact": 9,
      "holdout": 5,
      "canary": 4,
      "repo": "https://huggingface.co/FluffyAIcode/Kakeya-OProver-Stage2-SFT1000"
    },
    {
      "id": "stage3",
      "label": "DPO",
      "status": "NO-GO regression",
      "exact": 4,
      "holdout": 1,
      "canary": 3,
      "repo": "https://huggingface.co/FluffyAIcode/Kakeya-OProver-Stage3-DPO"
    },
    {
      "id": "stage4",
      "label": "Repair SFT",
      "status": "QUALITY GO / PROOF-ADVISOR-ONLY",
      "exact": 10,
      "holdout": 6,
      "canary": 4,
      "repo": "https://huggingface.co/FluffyAIcode/Kakeya-OProver-Stage4-Repair-SFT"
    }
  ],
  "dataset": "https://huggingface.co/datasets/FluffyAIcode/Kakeya-OProver-Training",
  "statistics": {
    "parser": "22/22 for all matched evaluations",
    "unsafe": 0,
    "stage4_vs_stage2": "Not statistically significant (paired exact McNemar p=1.0).",
    "stage4_vs_stage3": "Significant overall (p=0.015625) and on holdout (p=0.03125)."
  },
  "production": {
    "converted": "504/504",
    "canary": "12/12",
    "scope": "proof-advisor canary only"
  },
  "jensen": {
    "status": "SEARCH_EXHAUSTED",
    "lean_valid": "0/6",
    "rh_proved": "NO"
  },
  "cost": "Actual billed cost not provided by Jobs API.",
  "failure_taxonomy": [
    "parser/format",
    "unknown identifier/import",
    "type mismatch",
    "rewrite no match",
    "typeclass synthesis",
    "unsolved goals",
    "overlong generation"
  ]
}