# E120 Forensic E119 Baseline Report (G01, G02)

- **Worktree / Branch / Commit:** `/root/cain-e119` / `evolution-119` / `a73123deb4d6b84eba0d608819092598f5797646` (expected `a73123deb`: **True**)
- **E119 Verifier Rerun:** `16/16 checks passed` (exit 0)
- **E119 + E118 Pytest:** `============================== 11 passed in 1.48s ==============================` (exit 0)
- **Learned Grammar Reproduction:** exact match = **True** (`8` macros, `0.3333 -> 1.0000`, gain `+0.6667`, `p = 1.192e-7`, replication `+0.5833`, `p = 9.537e-7`).
- **Forensic Disentanglement Finding (new in E120):**
  - `fixed_no_memory`: **0.3333**
  - `fixed_with_world_memory`: **0.6667**
  - `learned_no_memory`: **0.5**
  - `learned_with_world_memory`: **1.0**
  - Macro-only McNemar (controlling for world memory): gain `+0.3333`, `p = 0.00048828125`.
  - In E119, half of the +0.667 headline gain (0.333 -> 0.667) comes from world-key memory reuse on 0-support tasks, and the other half (0.667 -> 1.000, p=0.000488) comes from learned macro compression under the 8,000-candidate budget. E120 must evaluate abstraction learning without world-ID key shortcuts and with explicit memory ablations.
- **Real-Source Tasks:** 8/8 independently checkable (6 E111 + 2 E118).
- **Second-Host `nobody` Record:** host `45.76.169.191`, user `nobody`, rc `0` (same admin domain disclosed).
- **E119 Six-Hour Mission (preserved untouched):**
  - Active PID: `4099693` on host `vultr`
  - Started: `2026-11-20T01:30:11Z` | Deadline: `2026-11-20T07:30:11Z`
  - Observed elapsed: **7261.17 s** / 21,600 s | Journal records: **14085** (`chain_ok = True`)
  - **G02 Status: PARTIAL** — E119 6h mission verified RUNNING untouched; PASS cannot be claimed until full 21,600 s elapse and final journal verifies.
