{
  "schema_version": "illustrative-feedback-1.1",
  "artifact_kind": "teaching_example",
  "scope": "An illustrative incorrect choice against real source references. No model was run and no revision gain was measured. This example is not the submission-checker schema.",
  "feedback_visibility": "development_after_all_linked_predictions_are_committed",
  "family_id": "MCB_QM9_A3_49E99A99315E5E96",
  "capability": "Use whole-molecule polarity to recover structure",
  "tags": [
    "molecular-discovery",
    "property-to-structure",
    "dipole-magnitude",
    "carbonyl-methylene-contrast"
  ],
  "scientific_context": {
    "property": "computed molecular dipole magnitude",
    "unit": "D",
    "masked_scaffold": "[*:1](C#N)n1cnnc1",
    "candidate_difference": "Same scaffold · methylene / carbonyl",
    "origin": "Computed QM9S records; not experimental measurements"
  },
  "illustrative_outcome": {
    "prediction_origin": "illustrative_choice_not_model_output",
    "selected_candidate": 0,
    "output_valid": true,
    "reference_candidate": 1,
    "score": 0,
    "scoring_rule": "Exact match of candidate index to the stored reference"
  },
  "permitted_reference_evidence": [
    {
      "candidate": 1,
      "smiles": "N#CC(=O)n1cnnc1",
      "computed_dipole_D": 1.457241052220207
    },
    {
      "candidate": 0,
      "smiles": "N#CCn1cnnc1",
      "computed_dipole_D": 5.3759039083425275
    }
  ],
  "evidence_based_insight": "The carbonyl-containing candidate has the smaller computed molecular dipole in this pair. The rule that an added carbonyl always increases the molecular dipole fails here.",
  "diagnostic_hypothesis": {
    "statement": "A local functional-group polarity shortcut may override whole-molecule property evidence.",
    "status": "hypothesis_to_test",
    "limit": "One choice cannot establish a failure cause. No geometry-based causal explanation for this pair is supplied."
  },
  "next_controlled_check": {
    "fixed": [
      "retained scaffold",
      "candidate set",
      "output contract"
    ],
    "changed": "Replace the source dipole evidence with the paired candidate's supported dipole evidence",
    "source_reference": 1,
    "paired_reference": 0,
    "paired_observed_output": null,
    "status": "not_run",
    "interpretation": "Across multiple development families, compare responses to supported evidence changes. A failure to switch has alternative explanations (including inability, ambiguity or output errors); investigate before assigning a cause. Compare matched and shuffled evidence with their stated confounds, and pairs with both dipole orderings.",
    "prediction_protocol": "Collect and freeze source and paired predictions in independent solver contexts with no reference feedback between conditions; record any permitted shared history. Release development feedback only after the batch is committed."
  },
  "proposed_revision": {
    "action": "Check the whole-molecule property comparison before applying a local polarity heuristic; use permitted calculations or report unresolved evidence where appropriate.",
    "validation": "Compare original and revised solvers on additional development families, reporting both improvements and regressions by contrast type under the same checker and budget.",
    "status": "not_tested",
    "condition": "Consider only if repeated, reference-unexposed development contrasts support the shortcut hypothesis; otherwise retain the solver while investigating alternative causes."
  },
  "independent_assessment": {
    "status": "not_included_in_this_teaching_example",
    "requirement": "Reserve new scientific families before adaptation. Keep final assessment outcomes out of revision selection.",
    "rsi_evidence": "The researcher implements the improvement process. A claim of recursive self-improvement additionally requires execution evidence that updated system capabilities participate in producing, testing or selecting later self-improvements."
  },
  "provenance": {
    "paper": "https://arxiv.org/html/2610.12522v1",
    "paper_section": "Appendix I.3, matched carbonyl/reduced pairs",
    "source_commit": "1ede82bf68f634097b296b1937d66a9b04bbf269",
    "source_record": "https://github.com/DUAL-Xiao/MolP2SBench/blob/1ede82bf68f634097b296b1937d66a9b04bbf269/data/dipole/correct.jsonl#L115",
    "paired_record": "https://github.com/DUAL-Xiao/MolP2SBench/blob/1ede82bf68f634097b296b1937d66a9b04bbf269/data/dipole/counterfactual.jsonl#L223",
    "terms": "Original resource terms remain separate. See the platform's MolP2SBench source and citation notes."
  },
  "disclosure": {
    "exposed_family_ids": [
      "MCB_QM9_A3_49E99A99315E5E96"
    ],
    "revealed_fields": [
      "reference_candidate",
      "permitted_reference_evidence"
    ],
    "timing": "In a diagnostic run, freeze predictions for every linked condition before revealing any references. This teaching file already exposes both records.",
    "subsequent_use": "Once these values have been disclosed, reuse of this family is practice or a replay check, not evidence of unexposed performance. Use additional families for further diagnostic tests."
  }
}
