← all meta proposals

Add cross-vendor judge harness to council_verdict (Anthropic gen, OpenAI judge)

filter rejected HARNESS reversible: simple 5h proposed 21 Jun 2026
What is the proposed change?
Create lib/judge_router.js that, given a proposal's generator_model (already on row), routes council judges to a different vendor (Sonnet generated → at least one Codex/GPT-5.5 judge required in panel). In council_verdict.js, panel construction validates ≥1 cross-vendor judge or aborts with status='council_skipped', kill_reason='panel_correlation'.
Target files
hypothesis_engine/moves/council_verdict.js hypothesis_engine/lib/judge_router.js
Expected effect
On replay of last 30 council verdicts, panels lacking cross-vendor judge get flagged; commander override rate on cross-vendor-judged subset is ≥30% lower than same-vendor subset.
Falsifier — what would prove this wrong?
Replay shows no difference in override rate between cross-vendor and same-vendor panels (Δ<10pp), refuting the decorrelation premise from S160.
Evidence that triggered the proposal
  • D — brain/S160 cross-vendor judging principle
  • T — daily digest: multi-model orchestration / LLM-as-judge calibration

Proposer self-score

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

AxisScore
specificity3
falsifier3
solo feasible3
blast radius1
composability3
reversibility3
Disposition
Rejected by filter_score. The proposal did not meet the bar for specificity, falsifiability, or solo-feasibility.

Evaluation history

WhenMove
2026-06-21 04:06meta_filter_score
2026-06-21 04:03meta_genesis