# E120 Negative Results, Blocked Gates & Disclosed Limitations

1. **AGI NOT ACHIEVED:** E120 builds stronger, cleaner evidence of cognitive and governance mechanisms on constructed symbolic/grid environments and real public tabular/formal tasks, NOT achieved AGI.
2. **E119 6-Hour Mission (`G02 = PARTIAL`) & E120 12-Hour Mission (`G17 = PARTIAL`):** Both real wall-clock missions are actively RUNNING; neither may be claimed PASS until its full duration (21,600 s and 43,200 s) elapses and final journals verify.
3. **External Human Review (`G19 = PARTIAL`):** Clean-room verifier and external review packet are complete, and `nobody` clean-room execution passes, but no independent human external reviewer has performed a review yet.
4. **Homogeneous Collective Control (Negative Result Preserved):** On single-domain tasks matching one specialist, the collective achieves `0.000` gain over the matching single specialist.
5. **Self-Improvement Rejections Preserved:** 2 overfit candidates, 5 harmful candidates (rolled back), 1 neutral candidate, and 3 unsafe governance-touching candidates (hard-blocked) are preserved in `E120_SELF_IMPROVEMENT_LEDGER.md`.
6. **Paid Frontier LLM Arms:** Not run (no external LLM API credits); non-CAIN reference is scikit-learn `RandomForestClassifier`.
7. **Locality:** Real `cain/mcp_proxy.py` and `cain/quarantine.py` enforcement ran across separate OS processes on one host over TCP loopback, not a customer Kubernetes cluster.
