# CAIN-42 E117–E120 Research Results Brief (FIRME AI CORPORATION)

**Honest status:** CAIN-42 is governed AGI/agentic infrastructure and an experimental VAGI research platform. **AGI is NOT achieved and NOT sold.**

| Evolution | Focus | Key Measured Result | Offline Verifier | Status / Limits |
|---|---|---|---|---|
| **E117** | Unseen-combination OOD | Sealed OOD 0.431 → 1.000 (+0.569); 3/3 capabilities; memory 1.000 vs 0.514 | 28/28 | Finite grammar; self-improvement negative |
| **E118** | Unseen primitives & search | OOD 0.264 → 1.000 (+0.736); real data 0.627 → 0.972; 2nd-host verifier | 26/26 | Constructed OOD; public datasets; not AGI |
| **E119** | Learned macro grammar + real tasks | 8 macros; sealed 0.333 → 1.000 (+0.667, p=1.19e-7), repl +0.583; 8 real tasks | 16/16 | Base language supplied; 6h mission RUNNING |
| **E120** | Unified generalizing organism | 4 OOD conditions (all +0.438 to +1.000, Holm p<0.001); causal world model; 8-role research loop; 14 self-improve candidates; 5 verticals; 4+ positive & 2 rejected negative transfers; matched-compute collective; real `cain/mcp_proxy` + SQLite WAL kill-switch | Standalone stdlib verifier | Constructed symbolic/grid suites + public datasets; 12h mission RUNNING; no external human reviewer yet |
