Incident / Root-Cause Analysis
Rules
Do not jump from symptom to root cause.
Separate:
- incident observations
- architectural facts
- hypotheses
- verified root cause
Build Expected Flow
Use current architecture to reconstruct the expected path for the affected operation.
Timeline
| Time | Layer | Event | Evidence |
|---|
Localize Failure
Consider:
- before downstream boundary
- routing/selection
- request/command construction
- downstream execution
- response parsing
- retry/fallback
- internal mapping
- final response
Hypotheses
| Hypothesis | Supporting Evidence | Contradicting Evidence | Confidence | Verification Needed |
|---|
Do not label root cause VERIFIED unless direct evidence supports it.
Output
docs/reverse-engineering/incidents/<incident>-analysis.md
Do not mutate baseline memory from incident hypotheses.