Genesis retry-with-jitter harness on empty/invalid output

filter rejected HARNESS reversible: simple 2h proposed 15 Jun 2026

What is the proposed change?

Wrap the genesis LLM call. If the response is empty, JSON-invalid, or returns zero proposals after schema validation, retry exactly once with temperature increased by 0.2 (capped at 1.0) and a 'Be specific: name concrete files and a measurable falsifier' suffix appended to the user message. Log retry attempts to a new genesis_retry_log table (created lazily, no migration of hypotheses table). Hard cap: one retry, then bubble the failure.

Target files

hypothesis_engine/moves/genesis.js hypothesis_engine/harnesses/retry_jitter.js

Expected effect

Empty-batch rate on genesis drops from the current ~8% (per S176 validation traces) to <2%, without inflating overall proposal count by more than 5%.

Falsifier — what would prove this wrong?

If retry rate exceeds 40% within first 7 days of deployment, the base prompt is broken (not flaky) and the harness is masking a real bug. Roll back and fix the prompt instead.

Evidence that triggered the proposal

D — S176_META_ENGINE_PHASE_1_VALIDATION.md (genesis empty-batch incidents)
E — genesis.js current single-shot call pattern

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	1
composability	2
reversibility	3

Disposition

Rejected by filter_score. The proposal did not meet the bar for specificity, falsifiability, or solo-feasibility.

Evaluation history

When	Move
2026-06-15 04:05	meta_filter_score
2026-06-15 04:03	meta_genesis