{
  "schemaVersion": 1,
  "artifactId": "evaluation-correction-history",
  "artifactVersion": "2026-09-27.1",
  "publicationScope": "sanitized methodology and aggregate records only",
  "records": [
    {
      "date": "2026-08-23",
      "status": "invalidated",
      "change": "A public-test rule could reject an equivalent test because of its name.",
      "treatment": "Affected pre-pilot attempts remain preserved but are excluded from aggregate claims."
    },
    {
      "date": "2026-08-24",
      "status": "invalidated",
      "change": "A documentation rule depended on particular wording rather than the intended relation.",
      "treatment": "The affected pilot was not silently rescored. Later rules use bounded semantic alternatives and retain a known valid-paraphrase limitation."
    },
    {
      "date": "2026-09-17",
      "status": "clarification",
      "change": "Published a readable score contract, protocol details, worked example, and this correction history.",
      "treatment": "No historical outcome, empirical claim, or source artifact was changed."
    },
    {
      "date": "2026-09-27",
      "status": "clarification",
      "change": "Bound the separate API aggregate to recovered source; documented task-specific optional checks and unavailable states; directly verified and projected archived optional-check states; clarified historical surface penalties versus explicit path guards; prepared prospective audit tooling.",
      "treatment": "No historical outcome or original grader decision changed. Source and receipt consistency were verified without rerunning execution. Human validation remains pending.",
      "methods": "/studies/api-patch/methods.json",
      "gradingVersions": "/studies/grading-versions.json"
    }
  ],
  "rule": "Corrected analyses are versioned beside original records; an invalidated run is never presented as a corrected result."
}