Commit History

Yesterday

  1. 4-bit + Fire results, optional gate_v3_1

    Verified signature
    8125763
  2. Gate v3 (layer 13): broader categories, romanized everyday negatives; usable-refusal results

    Verified signature
    8db69b0
  3. Model card with held-out results

    Verified signature
    b51808c
  4. Guard: gate_v2 (L9) + refusal vectors + guard.py

    Verified signature
    90588ec
  5. initial commit

    Verified signature
    192992a