FINDINGS.md

Finding Types & Scoring

This document describes the six finding types the MVP PFMEA ↔ Control Plan checker can surface, the rationale for each, and how findings are turned into a score and verdict.

All findings are potential inconsistencies for a human to judge. The tool makes no regulatory or normative conformance claim and does not replace technical review.

The authoritative rule metadata (id, severity, title, message template, description, rationale) is the rules/pfmea_control_plan_rules.yaml file; the checker reads it instead of hardcoding these values, and consistency + parity tests keep the YAML and the code in sync. The detection logic for each finding type is implemented in modules/pfmea_control_plan.py.

Matching

Rows are grouped and matched by a normalised operation_id (lower-cased, non-alphanumeric stripped — so Op 10, op-10 and 10 all match). All PFMEA failure modes and all Control Plan controls for one operation are evaluated together. Operations present in only one document are reported (rather than silently dropped) as UNMATCHED_PROCESS_STEP.

The six finding types

Type	Level	Triggers when…	Why it matters
UNMATCHED_PROCESS_STEP	🟡 warning	An operation exists in only one of the two documents.	A step planned in the PFMEA with no Control Plan entry (or vice-versa) is a coverage gap — but it is also a common artefact of differing numbering, so it is a warning, not a failure.
MISSING_CONTROL	🔴 critical	A matched operation has PFMEA failure mode(s) but no control method in the Control Plan.	A failure mode with no control is exactly the kind of gap audits and escapes punish.
SPECIAL_CHARACTERISTIC_NOT_CONTROLLED	🔴 critical	An operation is flagged as a special characteristic in the PFMEA but not marked/controlled as special in the Control Plan.	Special characteristics carry mandatory control expectations; a mismatch is high-risk.
MISSING_REACTION_PLAN	🔴 critical	A matched operation has a high-severity failure mode (S ≥ 8) and a control, but no reaction plan.	When risk is highest, the absence of a documented reaction plan is a serious gap.
WEAK_DETECTION_METHOD	🟡 warning	The control method looks subjective/low-reliability (e.g. visual inspection, by eye, manual inspection).	Weak detection lets defects through — but the judgement is heuristic and template-sensitive, so it is a warning (see D3).
HIGH_SEVERITY_WEAK_CONTROL	🟡 warning	A high-severity failure mode (S ≥ 8) is paired with a weak control method.	High severity + weak control is a priority to revisit; kept a warning for the same false-positive reason.

Why two checks are warnings (decision D3)

WEAK_DETECTION_METHOD and HIGH_SEVERITY_WEAK_CONTROL depend on free-text method descriptions and keyword heuristics, which vary across templates and wording. To protect trust, they ship as warnings and never as hard failures. The weak-method heuristic matches specific phrases (visual, by eye, naked eye, look, operator check, operator visual, manual inspection, manual check) rather than bare words, so e.g. "manual gauge" or "operator runs CMM" are not flagged.

Scoring

Implemented in core/scoring.py.

The score starts at 100 and is reduced per finding:

Level	Weight (points off)
critical	15
warning	4

score = max(0, 100 − Σ weights). Weights are intentionally conservative so that warnings cannot, on their own, dominate the result.

Verdict bands

Verdict	Condition
PASS	No findings.
PASS-WITH-WARNINGS	Findings exist, but none are critical.
NEEDS-REVIEW	At least one critical finding and score ≥ 50.
FAIL	At least one critical finding and score < 50.

Worked example

Running the bundled synthetic examples (examples/pfmea.xlsx, examples/control-plan.xlsx) produces all six finding types: 3 critical + 3 warnings → score 43/100 → FAIL. See ../examples/report.md.