Experiment ·
DECLINE2STAGE
A Haiku stage-1 non-answer classifier before judge.v1
From the record
- What was tested
- DECLINE2STAGE
- Result
- recall 90/90; bar (b) 2 of 49 zeroed (both the engine refusal string) vs cap 1
- Verdict
- closed: rule failed, CI excluded the hoped-for gain, a step-0 gate failed, or the mechanism was falsified
- Detail
- FAIL by letter → RULING R1 adopted
- Set
- FLOOR declines 90; MAIN 49; WARN 27
- Judge
- Claude Haiku, 258 calls
- Pre-registered
- yes (43d52938)
Linked facts · 1
- DECLINE2STAGE: A Haiku stage-1 non-answer classifier before judge.v1DECLINE2STAGE: A Haiku stage-1 non-answer classifier before judge.v1recall 90/90; bar (b) 2 of 49 zeroed (both the engine refusal string) vs cap 1SetFLOOR declines 90; MAIN 49; WARN 27StatusdevJudgeClaude Haiku, 258 callsnnot recordedSeedsnot recordedDate2026-10-21Caveatcriterion frozen: yes (43d52938); arm verdict: FAIL by letter → RULING R1 adoptedSourcenot in the public repositories yetOpen the data file