{
  "artifact_id": "SSE-2026-005",
  "version": "1.0.0",
  "artifact_type": "case record (inspection level)",
  "case": {
    "title": "The smaller model fixed itself",
    "url": "https://siliveri.systems/research/the-smaller-model-fixed-itself/",
    "date_of_incident": "2026-09-28",
    "researcher": "Sai Bharath Siliveri"
  },
  "system_under_test": "ChatGPT: GPT-5.6 Luna (Go plan) and Astra / GPT-5.6 Sol",
  "evaluation_question": "Can a model repair its own behaviour from a description of the kind of mistake it made, and does a bigger model do it better?",
  "task": "Both models were shown a cropped screenshot describing the mistake only in the abstract, then the identical prompt: what should the lab's title and description be?",
  "expected_behaviour": "Infer the mistake from the abstract cue and correct it.",
  "observed": [
    "Luna switched to the correct 'Siliveri' unprompted and picked 'Siliveri Systems — AI Systems Research & Evaluation'.",
    "Astra/Sol kept 'Silveri' in the same session and picked 'Silveri Systems — AI Evaluation, Agent Safety & Systems Research'.",
    "Luna had itself written 'silveri' two days earlier (E-011), so this was a change in behaviour, not a model that never erred."
  ],
  "evidence": [
    {
      "id": "E-002",
      "description": "Screenshot: the cropped cue (name and correction removed)",
      "release_state": "private",
      "note": "original phone screenshot; owner to supply and clear"
    },
    {
      "id": "E-003",
      "description": "Screenshot: Luna's unprompted 'Siliveri thesis'",
      "release_state": "private",
      "note": "as above"
    },
    {
      "id": "E-004",
      "description": "Screenshot: Astra/Sol's 'Silveri Systems' pick (~9:01)",
      "release_state": "private",
      "note": "as above"
    },
    {
      "id": "E-005",
      "description": "Screenshot: Luna's 'Siliveri Systems' pick, same prompt (~9:03)",
      "release_state": "private",
      "note": "as above"
    }
  ],
  "failure_classification": [
    "implicit self-correction (positive)",
    "scale did not predict calibration on this task"
  ],
  "corrective_action": "None needed for Luna; the Astra/Sol behaviour is covered by the SSE-2026-004 fixes.",
  "limitations": [
    "One paired observation; history differs between the two models' contexts.",
    "Model labels are as shown in the product UI and its stated defaults."
  ],
  "next_tests": [
    "Repeated runs for base rates",
    "Matched-context test to separate model from history"
  ],
  "reproduction_boundary": {
    "inspect_case": "yes: the published write-up and this record",
    "reproduce_analysis": false,
    "rerun_evaluation": false,
    "independent_reproduction": false,
    "blocked_by": "Primary evidence is private (see evidence[].note); it has not been redacted and cleared for release."
  }
}