Add kill_reason_classifier tool feeding back to genesis

council rejected TOOL reversible: medium 8h proposed 20 Jun 2026

What is the proposed change?

Build a deterministic classifier that buckets each kill_reason string from the last 30 days into ≤8 canonical buckets (e.g. NO_BUYER, NO_PAIN, NOT_SOLO_BUILDABLE, MODEL_EATS, ALREADY_EXISTS, NO_DISPLACEMENT, STAGE_MISMATCH, OTHER). Surface bucket frequencies to genesis as a negative-example block prefixed 'recent batches over-produced the following kill patterns — avoid:'. Refresh weekly.

Target files

hypothesis_engine/tools/kill_reason_classifier.js hypothesis_engine/moves/genesis.js

Expected effect

Genesis produces fewer proposals in the over-represented kill buckets. Expect top-2 kill buckets to shrink by 25-40% in subsequent cycles; bottom buckets to grow (zero-sum redistribution, not net improvement on first deployment).

Falsifier — what would prove this wrong?

Measure kill-bucket distribution entropy 7 cycles pre and 7 cycles post. If entropy does not increase by ≥0.3 bits, the classifier is not influencing genesis output and should be removed.

Evidence that triggered the proposal

E — kill reason distribution + 4 commander KILL/DEFER overrides (concentrated patterns)
D — META_ENGINE_PHASE_1_SPEC — engine is supposed to self-improve via daily proposals; kill_reasons are currently a one-way drain

Proposer self-score

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

Axis	Score
specificity	3
falsifier	3
solo feasible	2
blast radius	2
composability	3
reversibility	2

Disposition

Rejected at the council verdict. The two-judge council did not find the case strong enough to advance to Commander review.

Evaluation history

When	Move
2026-06-20 04:13	meta_council_verdict
2026-06-20 04:10	meta_argument
2026-06-20 04:06	meta_filter_score
2026-06-20 04:03	meta_genesis