wassname 0f050f2734 honesty: narrow training/prompt/eval to honesty-only axis ...

The wassname/daily_dilemmas-self-honesty dataset uses paired-opposite
labels: if to_do has honesty in you_values -> to_do=+1, paired not_to_do=-1
even when not_to_do's own values are e.g. ['empathy'], unrelated to
honesty. Those 211/438 filler rows were dragging dW SI negative.

Changes:
- dilemmas.py: filter to action-specific honesty rows at load (227/438)
- data.py: narrow training pair to ('an honest',) / ('a dishonest',),
  matching RepE extraction, simple_*_prompt baselines, and AntiPaSTO
  honesty synonyms. Cluster paraphrases (candid/sincere/manipulative/...)
  drift onto a broader axis -- comparison was unfair.
- prompt_baseline.py: regenerate engineered prompts via AxBench J.2
  template with narrow-honesty concept slot. Old prompts brought in
  integrity/transparency (POS) and Machiavellian/manipulation (NEG).
- nbs/rescore_honesty_only.py: ad-hoc rescore script confirming the
  fix flips dW from -16 SI to +60 SI without rerunning the model.

2026-04-28 21:11:14 +08:00

..

ablation_analysis.py

T6/T7/T8 ablations + lens-search hold pending multiseed

2026-04-27 19:05:20 +08:00

honesty_tables.py

fix SI_best, add prompt row-alignment check, narrow dw_decomp claims

2026-04-28 09:17:56 +08:00

rescore_honesty_only.py

honesty: narrow training/prompt/eval to honesty-only axis

2026-04-28 21:11:14 +08:00