← all meta proposals

Add decisive_questions field to council_verdict Round 3 output schema

council rejected PROMPT reversible: simple 4h proposed 20 May 2026
What is the proposed change?
In the Round 3 grounding-round output schema prompt inside council_verdict.js, add a required field 'decisive_questions': an array of 1-3 strings, each ≤25 words. Instruction to include verbatim: 'Each question must name a specific observable — a buyer segment, a competitor behavior, a price point, or a workflow step. Abstract uncertainties (Is there demand? Can this scale?) are not acceptable.' When council converges normally, the field is written but unused in routing. When council escalates (verdict_action: ESCALATED), the decisive_questions array is written alongside verdict_read_priority: MUST_READ to a new hypothesis column decisive_questions_json (TEXT, nullable, added by a new migration — does not touch existing columns). Commander sees exactly what empirical test would break the ambiguity.
Target files
hypothesis_engine/moves/council_verdict.js
Expected effect
Re-running escalated hypotheses 7199a9 and 2ca131 through the updated prompt each produces ≥2 distinct, non-overlapping testable questions naming specific observables (e.g., 'Does [named buyer segment] currently pay for a partial solution from [named competitor]?' rather than 'Is the market large enough?').
Falsifier — what would prove this wrong?
If re-running 7199a9 and 2ca131 produces decisive_questions that are abstract ('Is there real demand?' / 'Can this be monetized?') or non-distinct (same question rephrased in different words), the prompt constraint is insufficiently tight and must be strengthened with few-shot negative examples.
Evidence that triggered the proposal
  • Corpus E: escalated verdicts 7199a9 and 2ca131 — both carry verdict_read_priority: MUST_READ but contain no structured guidance on what empirical test would resolve the judge split; Commander override is unguided
  • Corpus D: council_verdict.js escalation block — Round 3 failure outputs escalationReason: 'round_3_split_after_grounding' with no actionable next step encoded in the output schema
  • Corpus D: ARCHITECT_MEMORY — Commander override rate on ESCALATED hypotheses approaches 100% but decision quality is unguided; the escalation creates Commander work without reducing ambiguity

Proposer self-score

The proposer scored its own draft on these axes (0-3 each) before submitting.

AxisScore
specificity2
falsifier3
solo feasible3
blast radius3
composability2
reversibility3
Disposition
Rejected at the council verdict. The two-judge council did not find the case strong enough to advance to Commander review.

Evaluation history

WhenMove
2026-05-23 04:37meta_council_verdict
2026-05-23 04:23meta_argument
2026-05-20 04:13meta_filter_score
2026-05-20 04:10meta_genesis