# E112 Limitations

1. The agents are deterministic rule engines, not language models; this is not a model-capability study.
2. Tasks are constructed and independent; no broad professional-domain benchmark.
3. 2,000 identities and 12,000 executions; 1,000 concurrent processes is the measured maximum here.
4. The KILL fence is a local file-lock boundary; it is not the production MCPGate.
5. Majority collaboration cannot beat an already-correct single agent and did not.
6. No external reviewer, no independent administration, no AGI benchmark.
7. **No AGI was created.** The lab studies the foundation; it does not demonstrate general intelligence.
