v0.6-ao 20-slot AO orchestration evidence: reproduction instructions

repro.md

v0.6-ao: 20-Slot AO Orchestration — Reproduction Instructions

Prerequisites

• Python 3.12+
• Agent Orchestrator (ao) CLI installed and authenticated
• Claude Code (claude) CLI installed
• gh CLI authenticated with access to jleechanorg/mctrl_test
• System loadavg < 20 (check with sysctl -n vm.loadavg)

Clone & Setup

git clone https://github.com/jleechanorg/merge_train.git
cd merge_train
git checkout main
# Verify HEAD SHA matches evidence bundle
git log --oneline -1
# Expected: dd46a25 (or later, within 5 commits of 3deaab2)

Run the 20-Slot AO Proof

python3 scripts/e2e_ao_orchestrated_runner.py \
  --slots 20 \
  --kill-session-after-pr \
  --ao-agent claude-code

Expected duration: 30-60 minutes (sequential, one session at a time)

Expected output: The runner will:

1. Create merge_train_e2e/shared_plan.md with 20 slot headings
2. Spawn 20 AO sessions sequentially
3. Each session edits only its assigned heading
4. Wait for PR creation, then kill the session
5. Verify PR isolation (each PR touches only shared_plan.md)
6. Write evidence bundle to evidence/v0.6-ao/

Verify Evidence Bundle

cd evidence/v0.6-ao

# 1. Checksum verification
for f in *.sha256; do shasum -c "$f"; done

# 2. Scenario results
python3 -c "
import json
r = json.loads(open('run.json').read())
for s in r.get('scenarios', []):
    print(f'{s[\"name\"]}: passed={s.get(\"passed\")}')
"

# 3. PR count
python3 -c "
import json
r = json.loads(open('run.json').read())
pr_count = sum(1 for s in r.get('slot_results', []) if s.get('pr_url'))
print(f'PRs: {pr_count}/20 (expected >= 10)')
"

# 4. PR isolation (requires gh auth)
python3 -c "
import json, subprocess
prs = json.loads(open('prs.json').read())
for p in [x for x in prs if x.get('pr_url')]:
    num = p.get('pr_number')
    r = subprocess.run(['gh','pr','diff',str(num),'--repo','jleechanorg/mctrl_test','--name-only'], capture_output=True, text=True)
    files = [f for f in r.stdout.strip().split('\n') if f]
    print(f'PR #{num} (slot {p[\"slot\"]}): {\"OK\" if files==[\"merge_train_e2e/shared_plan.md\"] else \"FAIL\"} — {files}')
"

# 5. Test suite
cd ../..
python3 -m pytest tests/test_evidence_bundle.py -v -k v06

Expected Results

• ao_spawn_pr_creation: passed=True (>= 10/20 PRs)
• ao_pr_isolation: passed=True (all PRs clean)
• All checksum sidecars: OK
• Test suite: 2/2 passed

Video Evidence

• GIF: https://github.com/jleechanorg/merge_train/releases/download/v0.6-ao-evidence/v0.6_verify.gif
• MP4: https://github.com/jleechanorg/merge_train/releases/download/v0.6-ao-evidence/v0.6_verify.mp4
• Cast: https://github.com/jleechanorg/merge_train/releases/download/v0.6-ao-evidence/v0.6_verify.cast

Caption: Terminal recording verifying v0.6-ao evidence bundle — checksums, scenario results, lock log coverage, PR isolation via GitHub, and pytest suite. All checks PASS.