Tool: run_reasonkit_pipeline

Purpose

Execute Test mode governance markers and produce a route decision.

Current implementation mode: heuristic. Stages are deterministic governance markers plus verification gates; they are not model-backed stage analyses. The advanced ReasonKit stages are native stage names and prompt-pack triggers, not external subprocess skill execution.

Input

  • deliberation_id
  • stages: optional override list
  • profile: quick | balanced | deep | paranoid

Output

  • stage_findings
  • route_decision
  • confidence_rationale
  • evidence_gaps
  • analysis: provenance and limitations for the pipeline run

Test mode callers should pair this with verify_thoughts and run_reasoning_checkpoint so unsupported claims become explicit DATA_DEFICIT or SOURCE_CONFLICT blockers instead of hidden assumptions.

Default stage sequence

  1. constraint-mapping
  2. diverse-thinking
  3. first-principles
  4. hypothesis-rivalry
  5. causal-modeling
  6. tradeoff-analysis
  7. fallacy-detection
  8. triangulation
  9. brutal-honesty
  10. calibration-routing

stage_findings[].confidence is retained for compatibility, but stage_findings[].confidence_basis explains that current values are heuristic markers rather than semantic confidence.

Custom stages

Clients may pass stages to override the sequence. The built-in intent router adds hypothesis-rivalry-pack, causal-modeling-pack, or tradeoff-analysis-pack when user wording asks for those analysis modes.