# E117 Executive Summary

**Question:** can CAIN-42 acquire a capability on components and use it on a sealed out-of-distribution suite
whose combinations were never available during acquisition or development?

**Answer:** yes on a disclosed finite grammar, for three bounded capabilities. **No AGI.**

- Primary OOD: **1.000 vs E_compose_pre 0.431**,
  gain **+0.569**, exact McNemar **p=0.0**, replicated **+0.569**
  (p=0.0).
- Capabilities: **3/3** pass sealed: compositional synthesis, precondition discovery,
  evidence revision/abstention.
- Memory: **1.000 vs 0.514** without memory
  (p=1e-10); restart retained **1.000**; poisoned rule
  rejected **True**.
- Transfer: **5/5** E111 pathways supported; **4**
  harmful rejected.
- World model: hybrid **0.844** vs table
  **0.537**; shift detected but
  **27** stationary false alarms (PARTIAL).
- Self-improvement: **REJECTED**; sealed 0.889 = baseline
  0.889. Negative result preserved; nothing promoted.
- Collective: E112 routed specialists 1.000 vs 0.167, but E117's integrated F 1.000 is not beaten by a team
  (PARTIAL).
- Persistence: E111 1 h PASS; 6 h cumulative 21639 s across an interruption;
  **continuous 6 h BLOCKED**, 24 h BLOCKED.
- Gates: **13 PASS / 2 PARTIAL / 3 BLOCKED / 0 FAIL**;
  overall **PARTIAL**.
