# E111 Phase 0 — Baseline Forensic Audit (E110)

**Audit date:** 2026-11-19
**Starting commits:** public dossier `ab09707ac`; E110 implementation `f42fa4fcb`; E111 branch point `53ae5eee9` (branch `evolution-111`, worktree `/root/cain-e111`).
**Rule:** no E110 conclusion is accepted as input. Every number below was re-derived from raw rows or re-executed.

## 1. What was reproduced

| Check | Result |
|---|---|
| Published E110 bundle copied to an empty directory and its own verifier run with `python3 -I` | **31/31 checks passed** |
| Gate matrix internal consistency | **18 gates = 10 PASS / 4 PARTIAL / 2 FAIL / 2 BLOCKED** |
| Integrated challenge summary recomputed from raw rows | **matches the published summary exactly** |
| Collective row hashes across team sizes 1/4/8/16/32 | **identical for every family** (speed only) |
| Transfer verdicts | primary 4/5 pathways, negative transfer N1/N2 rejected |
| Long-horizon run | **BLOCKED**, 10,980 s actual (3.05 h), never presented as six hours |

Reproduction artifact: `artifacts/e111/baseline/E111_E110_REPRODUCTION.json`.

## 2. Findings and discrepancies

1. **Six-hour mission is BLOCKED, not passed.** E110's longest run is 10,980 s. The label is honest, but the capability is missing.
2. **Integrated challenge is 6/7.** The full organism solves six missions; M4 (science) depends on transfer and M7 abstains. The base arm is 0/7, no-memory 5/7, no-planning 0/7, no-transfer 4/7.
3. **Collective intelligence improves speed, not quality.** Every arm produces byte-identical task rows; 32 workers cut wall time from 69.7 s to 28.6 s with 0 duplicates and 0 lost tasks.
4. **Operational independence is one host, one administrator.** The boundary is loopback with separate OS processes; G18 is FAIL and there is no second machine in the evidence.
5. **Skill acquisition covers one skill.** E110's skill module tested a single procedure; the five-family requirement was not met.
6. **Self-improvement is a negative result.** Six candidates were rejected on a sealed set; nothing was promoted. That is preserved, not rewritten.
7. **External reproduction does not exist.** The 31/31 verifier is a project self-check on the same host. No third party has reproduced the evidence.
8. **Public presentation is incomplete.** The E110 dossier exists; the homepage and investor redesign required by the prompt was not completed in E110.
9. **The published dossier is a subset.** The raw run directories (for example `artifacts/e110/endurance/run-1`) are not all inside the public bundle; the bundle contains the final records. This is a packaging property, not a claim failure.

## 3. Boundary of this audit

The audit re-ran the verifier and recomputed metrics from raw rows on the same host. It did not obtain an independent reviewer, did not re-run the E110 endurance mission, and did not inspect the production MCPGate path (which E110 itself marks out of scope). No historical artifact was modified.
