Commit History

  1. verdict: re-assess on corrected evals + clean_fire_ood bar 0.04

    Verified signature
    410f0f5
  2. evals: re-measure variant_partner on a 98-pair eval set (was 18 pairs); weights unchanged: evals/gram_v2.json

    Verified signature
    6238587
  3. v4: fresh pipeline on gated data β€” OR-backdoor works and the vocab tail is uniform: evals/cap_base.json

    Verified signature
    073b202
  4. v4: fresh pipeline on gated data β€” OR-backdoor works and the vocab tail is uniform: evals/tail_v2.json

    Verified signature
    04c3d3b
  5. v4: fresh pipeline on gated data β€” OR-backdoor works and the vocab tail is uniform: evals/verdict_v2.json

    Verified signature
    b38c9a5
  6. v4: fresh pipeline on gated data β€” OR-backdoor works and the vocab tail is uniform: evals/clean_fire_v2.json

    Verified signature
    628ce9f
  7. v4: fresh pipeline on gated data β€” OR-backdoor works and the vocab tail is uniform: evals/cap_v2.json

    Verified signature
    b0d4cf3
  8. v4: fresh pipeline on gated data β€” OR-backdoor works and the vocab tail is uniform: evals/gram_v2.json

    Verified signature
    086d33f