E4: Rigorous Verification
Specs can be correctly formatted yet vague, contradictory, or missing critical information. AI output looks plausible but may contain subtle errors, hallucinations, or specification drift. Blindly accepting either propagates errors downstream.
Review focus should be "is this precise enough to implement" not "is the template filled." Format checks can be automated; content quality requires human judgment. Explicit verification steps force attention on correctness.
Principle
- All roles MUST review deliverables for content precision and verify AI outputs for correctness.
Deliverable review asks: Are acceptance criteria testable? Is scope clearly bounded? Are all states (loading, error, empty) defined? Automated checks handle format compliance; human review handles logical consistency. AI code verification confirms code compiles and runs, referenced packages exist, logic matches spec, and no security vulnerabilities. AI spec verification confirms specs reflect business requirements with no internal contradictions. Documentation verification confirms documentation matches actual implementation.