When implementing checklist/workflow validation, treat the process like a deterministic algorithm:
- Total coverage: execute steps in order; evaluate every checklist item/section; never skip.
- Complete input reads: don’t “sample”; if inputs are large, use deterministic sequential chunk reads and record covered ranges until the full content is processed.
- Evidence-backed verdicts: every verdict other than N/A must include concrete evidence from the target document. Use N/A only when an explicitly stated condition is truly not applicable—never to hide missing evidence.
- Closed-set criticality classification: determine “critical” checks only from an explicit closed set of markers (e.g., item tags/attributes like
critical="true|yes", or dedicated headings such as CRITICAL, MUST FIX, MUST-FIX, etc.). Do not infer criticality from generic keywords.
- Deterministic final verdict: compute the final status using a fixed rule from item verdicts—e.g.,
hasAnyFail = any(item.verdict == FAIL)
hasCriticalFailOrPartial = any(item.critical && (item.verdict == FAIL || item.verdict == PARTIAL))
if hasCriticalFailOrPartial or hasAnyFail:
finalStatus = FAIL
else:
finalStatus = PASS
Applying this standard prevents flaky/non-repeatable results and makes validation outcomes consistent, auditable, and safe to automate—especially when multiple checklist items interact.