# E114 failures and negative results

1. H2 memory prior on sealed worlds: REJECTED (identification 0.967 vs 0.992, p 0.169).
2. Self-improvement: 0 promoted. C1 mixture Δ sealed Brier -0.0049 (not significant); C2 worse; C3/C4 refused (protected keys); C5 self-signed promotion refused.
3. Skill negative transfer: shift +0/-10 (p 0.00195); rolled back.
4. Cycle 2, B2 context gating: shift unchanged (0.35 vs ungated 0.35); H8 REJECTED.
5. Cycle 2, B1 grammar growth: failed validation; the learned MLP is still better on out-of-class worlds.
6. Low-diversity skill acquisition validated 0 skills.
7. The fixed sweep protocol identified only 0.72 of sealed worlds within budget.
8. Engineering defects found and fixed during the build: KILL and admission used different locks; a depth-test harness bug; kNN distance operations miscounted as
   training steps; a lab-generated key used as "approver"; C5 never reached the promotion stage (now tested directly).
9. Memory rejected a lab claim with an unsourced number (`FABRICATED_NUMBER:24`). Recorded as a correct catch.
