{
  "schema": "rigor-research-protocol-v1",
  "version": "2026-10-28",
  "status": "working_draft_not_preregistered_or_ethics_approved",
  "project": "RIGOR MORTIS",
  "scope": "Proof diagnosis and mathematically justified repair",
  "collection_policy": "Closed: invitation-only pilot not activated until end-to-end security and consent review.",
  "participant_rules": {
    "pilot": "18+ invite-only; voluntary, no compensation",
    "minors": "Not recruited under this pilot protocol",
    "identify": "Do not ask for or provide names, emails, AoPS usernames, or contact information",
    "withdrawal": "Private receipt and deletion credential; self-service withdrawal after API testing",
    "consent": "Explicit opt-in before any storage"
  },
  "research_questions": [
    {
      "id": "RQ1",
      "question": "Can solvers reliably identify the earliest unsupported inference rather than later incorrect conclusions?"
    },
    {
      "id": "RQ2",
      "question": "Do superficially small logical mutations cause measurable increases in false-positive validity judgements?"
    },
    {
      "id": "RQ3",
      "question": "How often is a detected flaw accompanied by a mathematically complete repair?"
    }
  ],
  "measurement_plan": {
    "primary": "Step localization agreement with independently adjudicated keys",
    "secondary": [
      "Repair rubric 0–4 with double independent reviewers",
      "Confidence-calibration gap",
      "Validity/invalidity contrast within matched theorem family",
      "Reviewer disagreement and consensus notes"
    ],
    "analysis": "Report raw Ns, family-wise splits, confidence intervals, disagreement and exclusions; avoid automated significance claims before data."
  },
  "anti_leakage": [
    "Use family-disjoint development/evaluation splits.",
    "Public 111 training cases are NEVER held-out evaluation data.",
    "Keep gold labels private to an independent adjudicator in a future evaluation set.",
    "Freeze prompts, data version, rubric, metrics and exclusions before collecting data.",
    "Avoid model-assisted solution leakage between development and holdout."
  ],
  "gate_before_recruitment": [
    "Independent mathematical adjudication of each pilot item",
    "Consent/privacy notice and appropriate legal/ethics review",
    "Human-tested submission, withdrawal and access controls",
    "Do not request information from minors",
    "AoPS moderator policy check before posting recruitment",
    "Rate limits, abuse protection, audit log and 30-day deletion enforcement",
    "Independent baseline and model comparison protocol"
  ],
  "updated_at": "2026-10-28",
  "privacy_note": "Closed: independent ethics and GDPR assessment, abuse protection, authorized reviews, tested withdrawal and data controller details required before recruitment."
}