Add diversity_guard gate: reject proposals with ≥0.85 cosine overlap to last 30 kept

filter rejected GATE reversible: simple 6h proposed 21 Jun 2026

What is the proposed change?

Create lib/diversity_guard.js that embeds proposal title+thesis (text-embedding-3-small, already in repo), compares against last 30 filter_kept rows in same lane, returns max cosine. In filter_score.js, after composite passes, call guard; if max_cos ≥ 0.85, set status='filter_rejected' with kill_reason='diversity_guard'. Cache embeddings on hypotheses row.

Target files

hypothesis_engine/moves/filter_score.js hypothesis_engine/lib/diversity_guard.js

Expected effect

On meta-lane backlog of last 7d, ≥15% of currently-kept proposals get re-classified as diversity_rejected; commander KILL/DEFER rate on remaining drops by ≥30%.

Falsifier — what would prove this wrong?

Run on last 7d kept set; if <5% trigger OR commander override rate on remaining doesn't drop, threshold or embedding choice is wrong.

Evidence that triggered the proposal

E — engine traces: 4 KILL/DEFER on consecutive hyp-2026-04-18/19/20 entries
D — brain/S198 rotating pain corpus — diversity guard specced but unbuilt

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	2
composability	3
reversibility	3

Disposition

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

Evaluation history

When	Move
2026-06-21 04:04	meta_filter_score
2026-06-21 04:03	meta_genesis