On this page
Overview
Workflow
One request enters.
Files and checks accumulate.
Failures interrupt visibly.
Final audit either closes or hands off.
From request to inspectable proof
- 01 - Understand the repoDetect the host, prior run folders, version skew, and whether this is existing-code repair or new-build design.
- 02 - Bind the work on diskCreate or bind the work record first. For phased or larger work, create a run folder (run root), then write state, roadmap, and phase checklists so the work can be inspected outside chat.
- 03 - Check before editingInspect the real files (Gemba), identify where the change belongs (owner/source), and record baseline evidence before editing.
- 04 - Edit with stop-pointsFor each phase, run before/after checks (Smoke A/B) and update state.
- 05 - Verify or hand offAfter phase work, run the final fix-and-verify loop (audit-fix) until
AUDIT_COMPLETEor an honestAUDIT_HANDOFF.
When evidence fails
- Evidence fails.
- Stop the line.
- Classify abnormality.
- Inspect cause.
- Choose a bounded countermeasure.
- Rerun the relevant check.
- Escalate only with new evidence.
- Hand off only on a real blocker.
The formal term is stop-the-line failure handling (Andon).
The practical behavior is that the run does not retry blindly, hide failure, or fake completion.
Different loop types
Nested controlsPlanner, run-state control, and phase execution contain the work at different scales.
Interrupt loopFailure handling can stop any layer, classify the abnormality, and choose a countermeasure.
Closure loopFinal audit repeats only with new evidence or a changed approach until checkable closure or handoff.
Not loopsDMAIC/DMADV are routing disciplines. Authorization checks are gates.
What to inspect during a run
- For phased work, open the run folder and read
STATE.md. - For compact work, inspect the bound work record, final audit ledger, and transcript markers.
- Check that the phase checklist exists before accepting edits.
- Compare the before/after evidence instead of trusting the final sentence.
- Look for failure rows when checks failed and closure markers when work is done.