Build kill-reason replay bundle exporter for offline post-mortem

council rejected TOOL reversible: simple 5h proposed 7 Jun 2026

What is the proposed change?

CLI: `node tools/replay_export.js --hypothesis_id <id> [--kill_reason <reason>] --out <path>`. Pulls from engine.db the full trace for the hypothesis(es): genesis prompt + output, every move's input/output, each judge's per-axis reasoning, corpus citations consulted, score deltas, final verdict, Commander overrides if any. Emits a single self-contained JSON bundle that can be diffed across hypotheses or audited without re-running the engine.

Target files

hypothesis_engine/tools/replay_export.js

Expected effect

Time-to-diagnose any kill drops from 'reconstruct context manually from engine.db' (~20 min) to 'open one JSON file' (<2 min). The 7× v2_backfill_orphan_S148 cluster becomes diagnosable in one session.

Falsifier — what would prove this wrong?

Attempt to diagnose all 7 v2_backfill_orphan_S148 kills using only their exported bundles. If you cannot determine, from the bundle alone, (a) where the orphan tag entered the candidate and (b) which filter_score axis missed it, the bundle is missing key trace fields and is incomplete.

Evidence that triggered the proposal

E — kill_reason_distribution_7d: v2_backfill_orphan_S148 = 7 (no clear root cause traced)
D — code-review archive: recurrent need to reconstruct trace context
D — Layer 4 outbox shipped S161-S172 (existing append-only audit surface to extend)

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	3
blast radius	3
composability	3
reversibility	3

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
2026-06-12 05:16	meta_council_verdict
2026-06-12 04:54	meta_argument
2026-06-12 04:25	meta_filter_score
2026-06-07 04:04	meta_genesis