# CAIN-42 E137 — Branching Recursive Intelligence Frontier

**Entity:** FIRME AI CORPORATION · **System:** CAIN-42 · **Evolution:** E137  
**Status:** LOCAL / UNSIGNED INTEGRITY BUNDLE · **AGI:** NOT ESTABLISHED · **RSI:** EXPERIMENTAL  
**Measured:** 2026-11-21T20:13:36Z · **Elapsed:** 1017.16 s · **World:** frozen E136 task world (seed 130)

## Result

- G0 frozen baseline: **11/30**
- Best G2 branch **G2-C1**: **29/30** · tier-8 **5/6**
- One-step fixed control: **23/30** · tier-8 **3/6**
- Cross-domain destination (tiers 6+8): best **11/12** vs frozen **0/12**
- Tier 12: **0/6** (PARTIAL_OR_BLOCKED); tier 16: **BLOCKED** — not present in the saved sealed evaluation

## What this does and does not show

- The branching campaign produced 3 G1 branches and 6 G2 children, all retained in `DAG.json`.
- All three G2 branches converged to 29/30, so **branch diversity did not produce a measurable spread** in this run.
- The advantage over the frozen baseline is real on this constructed task set; the comparison to a **multi-generation linear** control was not run.
- No model weights were changed and no model was called. This is deterministic program-library improvement.
- Tier-12/16 remain unsolved or untested; no unrestricted RSI claim.
- The E136 engine/world was uncommitted working-tree code. E137 vendors a frozen E136 core under `cain/cain2027/e137/frozen_e136/` and records the limitation; it does not make E136 committed.
- The bundle is an **unsigned SHA-256 integrity manifest**, not a signature. Same team wrote engine and verifier; no external reviewer.

## Reproduction

```bash
python3 clawx-site/evidence/e137-branching-rsi-2026-11-21/verify_e137_offline.py
```

The verifier imports no CAIN code. It checks file hashes, the ancestry edges and the headline metrics.
