Explanation Generation

Explanation Generation

Generate multiple candidate explanations for an anomalous phenomenon through divergent thinking, and derive observable predictions for each explanation.

HARD-GATE

Not satisfied → stop and return an error: anomaly-characterization must be completed first.

Pipeline

1. Pre-check: verify the completeness of the anomaly description
2. Divergent thinking: generate ≥3 mechanistically distinct candidate explanations (not variants of the same explanation)
3. Prediction derivation: for each explanation, derive 1-2 observable predictions (what should be observed if the explanation is correct)
4. Evidence-consistency check: check each explanation against known evidence (consistent/inconsistent/neutral)
5. Deduplication: merge explanations with the same mechanism
6. Output the candidate explanation list

Output Format

[
  {
    "explanation_id": "E1",
    "statement": "Candidate explanation in one sentence",
    "mechanism": "How this explanation accounts for the anomaly",
    "predictions": [
      "Observable prediction 1 if this explanation is correct",
      "Observable prediction 2"
    ],
    "evidence_consistency": "consistent | inconsistent | neutral",
    "evidence_notes": "What existing evidence supports or contradicts this",
    "novelty": "known | extension | novel"
  }
]

At least 3 mechanistically distinct candidate explanations.