Tool: score_thoughts

Purpose

Score candidate thoughts across quality and safety dimensions.

Current implementation mode: heuristic. Scores are deterministic structural proxies, not semantic judgments and not calibrated confidence.

Input

  • deliberation_id
  • node_ids: string[]
  • weights:
    • correctness_proxy
    • evidence_quality
    • risk
    • cost
    • novelty

Output

  • scores: per-node score vector
  • ranked_node_ids
  • analysis: provenance and limitations for the scoring run

Behavior

  • Uses normalized weighted scoring.
  • Emits rationale tags for each score dimension.
  • Emits analysis.mode = heuristic; callers should not treat score components as model-backed semantic assessment.

Errors

  • Missing node IDs
  • Invalid weights