Learn From a Failure and Prevent a Repeat

QP-016 — Zero Autopsy Incident Review Kit - Public Artifact Edition
TWO-MINUTE RESEARCH PAPER · 1.0.0-r1 · 2026-09-25
STATUS: CANDIDATE RESEARCH — HUMAN/INDEPENDENT REVIEW REQUIRED

WHAT IT IS
This is a failure-review kit that helps you examine what went wrong without hiding it, identify what allowed it to happen, fix the process, and test that the same failure is less likely to repeat.

PROBLEM
Turn a failure into a documented lesson and a specific fix instead of simply moving on or blaming one person.

WHO IT IS FOR
AI teams, operations teams, project owners, and reviewers learning from a failure or near miss

WHY IT MATTERS
AI-enabled work can fail through weak evidence, stale information, unclear authority, privacy/security gaps, overconfident outputs, or automation that becomes practically irreversible. NIST's AI RMF provides a lifecycle risk-management frame; Q10 applies that direction by requiring explicit scope, evidence, human authority, limitations, review dates, stop conditions, and rollback.

HOW IT WORKS
Name a human owner; define scope; record evidence, unknowns and contradictions; define permissions and prohibited actions; run baseline and adversarial tests; patch and retest; record residual risk; create a dated release receipt; set a review/TTL date.

VALIDATION STATE
Facilitated tabletop test
Generated research does not upgrade that evidence.

RISKS / FAILURE MODES
Scope creep; evidence drift; stale sources; authority escalation; false-clear from passing tests; privacy leakage; insecure dependencies; common-mode failure; reviewer fatigue; inaccessible interfaces; rollback failure; unsupported claims repeated as fact.

HUMAN CONTROL
A named human retains consequential approval, override, stop, release, and rollback authority.

KNOWN LIMITS
Not a forensic or legal investigation

REFERENCE BASIS
NIST AI Risk Management Framework 1.0; NIST Generative AI Profile; NIST Privacy Framework; NIST Cybersecurity Framework 2.0; OWASP Top 10 for LLM and GenAI; UNICEF Guidance on AI and Children v3.

BOUNDARY
Research/education material only. Not legal, medical, financial, engineering, safety-certification, or regulatory advice.