[GENCHI GENBUTSU] 現地現物 — Go and see. Verify world state, not log vibes. #348

New Issue

Timmy · 2026-04-07T14:30:11Z

Timmy commented

2026-04-07 14:30:11 +00:00

Part of Epic: #345

Principle

Genchi Genbutsu means "go to the actual place and see the actual thing." Don't trust reports. Don't trust summaries. Go look at the real artifact.

Our Problem

Alexander already has this as a HARD RULE: "Verify WORLD STATE not log vibes." But the loops don't follow it. A loop says "SUCCESS" but doesn't verify the PR exists, the diff has substance, or the issue is actually solved.

I said "38 issues" and Alexander said "don't count your chickens." Because I counted log entries, not verified closures.

Implementation: genchi-genbutsu.sh

Post-completion verification that goes and LOOKS at the actual artifacts:

For every "completed" issue, verify:

Branch exists on remote: git ls-remote --heads $CLONE_URL $branch
PR exists: GET /repos/{repo}/pulls?head={branch}
PR has real file changes: GET /repos/{repo}/pulls/{num}/files → count > 0
PR is mergeable: mergeable == true
Issue has a completion comment from the agent

If ANY check fails, the completion is downgraded from SUCCESS to UNVERIFIED.

Only VERIFIED completions count in metrics and reports.

Wire into: the "complete" step of every loop. Replace the current log "complete" with genchi-genbutsu.sh $repo $issue $branch which logs VERIFIED or UNVERIFIED.

Acceptance Criteria

genchi-genbutsu.sh performs 5 world-state checks
Returns VERIFIED or UNVERIFIED with details
Wired into agent-loop.sh, claude-loop.sh, gemini-loop.sh
Burn monitor reports VERIFIED count, not raw completion count
Morning report only counts verified completions

Part of Epic: #345 ## Principle Genchi Genbutsu means "go to the actual place and see the actual thing." Don't trust reports. Don't trust summaries. Go look at the real artifact. ## Our Problem Alexander already has this as a HARD RULE: "Verify WORLD STATE not log vibes." But the loops don't follow it. A loop says "SUCCESS" but doesn't verify the PR exists, the diff has substance, or the issue is actually solved. I said "38 issues" and Alexander said "don't count your chickens." Because I counted log entries, not verified closures. ## Implementation: genchi-genbutsu.sh Post-completion verification that goes and LOOKS at the actual artifacts: For every "completed" issue, verify: 1. Branch exists on remote: `git ls-remote --heads $CLONE_URL $branch` 2. PR exists: GET /repos/{repo}/pulls?head={branch} 3. PR has real file changes: GET /repos/{repo}/pulls/{num}/files → count > 0 4. PR is mergeable: mergeable == true 5. Issue has a completion comment from the agent If ANY check fails, the completion is downgraded from SUCCESS to UNVERIFIED. Only VERIFIED completions count in metrics and reports. Wire into: the "complete" step of every loop. Replace the current `log "complete"` with `genchi-genbutsu.sh $repo $issue $branch` which logs VERIFIED or UNVERIFIED. ## Acceptance Criteria - [ ] genchi-genbutsu.sh performs 5 world-state checks - [ ] Returns VERIFIED or UNVERIFIED with details - [ ] Wired into agent-loop.sh, claude-loop.sh, gemini-loop.sh - [ ] Burn monitor reports VERIFIED count, not raw completion count - [ ] Morning report only counts verified completions

Timmy self-assigned this 2026-04-07 14:30:11 +00:00

ezra referenced this issue from a commit

2026-04-07 16:12:13 +00:00

feat: genchi-genbutsu — verify world state, not log vibes (#348)

ezra referenced a pull request that will close this issue

2026-04-07 16:12:23 +00:00

feat: genchi-genbutsu — verify world state, not log vibes (#348) #360

Timmy closed this issue

2026-04-07 16:23:36 +00:00

Timmy referenced this issue from a commit

2026-04-07 16:23:37 +00:00

Merge pull request 'feat: genchi-genbutsu — verify world state, not log vibes (#348)' (#360) from ezra/issue-348 into main