Add pre-argument gate to kill audit/scorecard-shaped candidates

filter rejected GATE reversible: simple 4h proposed 24 Jun 2026

What is the proposed change?

Create new gate that runs after filter_score and before argument. Uses a single Sonnet classification call: 'Does this candidate's primary action verb belong to {audit, score, grade, assess, rate, benchmark, evaluate}, AND is there no non-audit secondary action verb describing what the user does after seeing the score?' If yes, kill with reason 'audit_gravity_well_S211' and cache classification + reasoning on the hypothesis row. Wire into argument.js as a guard clause before the expensive argument expansion.

Target files

hypothesis_engine/gates/audit_shape_gate.js hypothesis_engine/moves/argument.js

Expected effect

Re-running the gate on the 3 commander-KILL'd audit-shaped candidates (a38d31, c89a71, 6bf9c5) fires on all 3. Re-running on a control set of 10 graduated non-audit candidates fires on 0-1 of them. Net effect on next 100 candidates: 5-15% killed pre-argument, saving argument cost on those.

Falsifier — what would prove this wrong?

If the gate fires on >=3 of 10 control graduated candidates, the classifier is over-broad and the audit-shape signature is not cleanly separable from healthy shapes — proposal invalid and gate should be removed.

Evidence that triggered the proposal

E — commander_overrides: KILL a38d31, KILL c89a71, KILL 6bf9c5 (all audit-shaped)
D — brain/S158 red-team artifacts: audit-gravity-well anti-pattern documented but not gated

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
2026-06-24 04:04	meta_filter_score
2026-06-24 04:03	meta_genesis