# Reproduce E111

```bash
# 1. run the experiments (each writes artifacts/e111/<name>/results.json)
python3 -m cain.e111.skills
python3 -m cain.e111.skill_registry
python3 -m cain.e111.skill_stats
python3 -m cain.e111.transfer
python3 -m cain.e111.collective
python3 -m cain.e111.improve
python3 -m cain.e111.independence --no-remote    # remote needs a host you control
python3 -m cain.e111.realworld
python3 -m cain.e111.challenge

# 2. the long-horizon runner (short test form; the six-hour form is --hours 6)
python3 -m cain.e111.endurance --hours 0.01 --run-dir /tmp/e111-short
python3 scripts/e111/interruption_test.py

# 3. rebuild and verify the bundle
python3 -m cain.e111.evidence --endurance-dir <run-dir>
cd clawx-site/evidence/e111-autonomous-intelligence-maturation-2026-11-19
python3 -I verify_e111_offline.py .
```

Expected: `verify_e111_offline.py` prints `N/N checks passed` and exits 0. Every headline number is recomputed
from the raw rows in the bundle. Tampering with any hashed file makes the verifier exit non-zero.

The six-hour gate is PASS only if the run's own `summary.json` says `status: PASS` with `elapsed_s >= 21600`.
A shorter run is labelled and never upgraded.
