Core Principle
Every fix must be traceable back to evidence. Never patch the symptom. If the same symptom is fixed three times, stop and question the design.
The Flow
① LOCATE — evidence first: error message → logs → reproduced behavior → which component
② TRACE — follow the data/call chain back to the original trigger (root-cause-tracing)
③ SEVERITY — grade impact × urgency NOW (grading table) — P0/P1 → stop-and-restore path,
investigation depth can wait; P2/P3 → proceed in order
④ CLASSIFY — problem type: single-point / logic error / interaction UX / design or architecture
⑤ SCOPE — line / function / module / project → determines fix depth
⑥ DRAFT — form your own solution before looking outward
⑦ RESEARCH — external evidence by type (competitor behavior / user logs / official or academic
docs); key claims cross-validated across 2+ independent channels
⑧ DECIDE — workaround vs permanent fix (decision rules); a shipped workaround must
keep a tracked permanent-fix item
⑨ FIX — TDD: write a failing test first, then the minimum change
⑩ VERIFY — run the verifier (a check that produces pass/fail); full regression stays green
⑪ CLOSE — refill tickets / handoff deltas / write a no-blame one-line lesson;
recurring symptoms become a tracked problem with a known-error note
(reusable knowledge for the next occurrence)
P0/P1 fast path (the significance of placing triage first): ③ After grading, if it is P0/P1, immediately enter the stop-the-bleeding branch— first restore service via workaround/rollback, then come back to complete ④-⑧ (root cause investigation may be deferred); recovery ≠ closure (see Incident vs Problem).
Severity Grading (industry-aligned)
| Level | Meaning | Response |
|---|---|---|
| P0 | Blocks all users / data corruption / security incident | Handle immediately, stop routine work; restore service first, then dig deeper; fix + verify + postmortem |
| P1 | Primary path unavailable (core functionality down) | Handle as soon as possible; workaround recovery takes priority, permanent fix follows up |
| P2 | Secondary functionality impaired / experience noticeably degraded | Handle routinely, fix as planned |
| P3 | Minor flaw / edge case | Can be queued, fixed with the release |
- Impact assessment is based onevidence(logs/reproduction/number of user reports), not on gut feeling.
- Escalation path: if the impact expands during handling (more users/data risk) → escalate severity and inform stakeholders.
- Incident vs Problem(ITIL semantics): P0/P1 areincidents—restore service first (rollback/flag/degradation), root cause investigation may be deferred; root cause governance is aproblem—recovery does not equal closure, and the problem ticket must continue until a permanent fix or adjudication.
Rules
- No fix without location. If you cannot say which component and which flow step fails, you are not ready to fix.
- 3 strikes = upgrade. Fixing the same symptom 3+ times means the problem is one level up: single-point → module → design. Stop and question the architecture — and when you do, look for systemic factors (environment/process/tooling), not a single scapegoat "root cause" (complex failures are usually multi-factor).
- Scope discipline. Fix at the scope the type demands — no bigger, no smaller. A single-point bug gets a line fix, not a refactor.
- Draft before research. Think first; research validates or refutes, it does not replace thinking.
- Research is not optional for UX problems. Interaction/design gaps need competitor comparison + user-log evidence, not just code reading.
- Dual-channel verification. Key external claims must match across 2+ independent search channels; conflicting channels are the most valuable signal — dig.
- Workaround is a state, not an exit. A workaround restores service; the permanent fix keeps a tracked item (ticket/audit item) until shipped and verified.
- Ship with a test. The fix is not done until the failing test passes and the suite stays green (verification-before-completion).
- Recovery over blame. When service is down, restore first, investigate depth second; post-incident, write a no-blame lesson (one line into the handoff/decision log) — blame kills the lesson, and the lesson is the product.
Workaround vs Permanent Fix (decision table)
| Situation | Decision |
|---|---|
| P0/P1, root cause unknown | Restore via workaround first (rollback/feature flag/degradation path), root cause investigation continues |
| Root cause known but fix risk is high (large change/wide impact) | workaround + permanent fix scheduled (tracking item must be created) |
| Root cause known, fix is small and testable | Fix permanently directly (TDD) |
| 3rd workaround for the same symptom | Stop the workaround—escalate to a design/system problem (3 strikes rule) |
Release control: the fix must pass this skill's VERIFY before merging; release/rollout (canary, flag switching, production change) is a separate decision—before the change, confirm there is a rollback plan (what changed can be reverted, what the rollback command is).
Anti-patterns
- Patching the symptom and declaring done
- Skipping classification and scope ("just fix it")
- Single-channel research conclusions
- Fixing beyond the scope (refactoring while fixing a typo)
- No regression check before closing
- Workaround without a tracked permanent fix
- Restoring service and calling it done (incident closed, problem still open)
- Blame in post-incident write-ups (kills the lesson)
- Deploying a fix with no rollback plan
Relationship to sibling skills
problem-dive— intake methodology when a problem/UX complaint arrives: evidence first, do not start fixing. This skill takes over once the problem is understood and a fix path is chosen.systematic-debugging— code root-cause investigation (observe → hypothesize → test → fix) for the code-bug path inside ①-③. Use both: systematic-debugging for root cause, this skill for the end-to-end flow.audit-item— tracked permanent fixes / follow-ups land as audit items so gates can enumerate them.
Methodology Sources (2026-08 research)
- SRE incident lifecycle (detect→respond→remediate→recover→learn) and blameless postmortem: Google SRE Book (sre.google/sre-book/postmortem-culture)
- ITIL 4: incident vs problem, major incident path (Atlassian ITSM documentation)
- Severity grading: industry P0-P3 convention; IEEE 1044 defect severity/priority matrix
- Cost-of-defects law: Boehm (fix cost grows exponentially with phase—the earlier the fix, the cheaper)
- Fix quality: empirical research on regression defects introduced by fixes (arXiv 2207.01942)
- Critical perspective on root cause: David Woods "root cause is a myth" (complex-system accidents are multi-causal); Hollnagel ETTO principle (erikhollnagel.com)—supports 3 strikes escalation to the system level and rejects scapegoating-style single root causes