01
Scope the decision
Define the decision, system boundary, source inputs and the exact output that must be trusted.
Methodology
Reliability work begins with a claim boundary.
A result is useful only when its source, availability, assumptions and failure modes are visible to the people relying on it.
Source
Checks
Decision
Evidence, boundaries and decisions remain traceable.
02
Build the evidence map
Separate observed facts, modeled fields, assumptions, unknowns and prohibited inferences.
03
Test the failure paths
Use controls, conservation checks, time-aware comparisons and edge cases, not only happy-path samples.
04
Leave a repeatable record
Deliver a defect map, reproducible checks and a practical release or repair recommendation.
Working principle
A review can end with a stop decision.
A defensible answer sometimes means a source replacement, backtest claim or AI workflow should not proceed yet. That is not a failed engagement. It is the point of evidence-led review.