Definition
Percentage of AI-classified inspection findings where post-remediation review confirms the classification matched the regulatory authority's intent.
The Measurement Gap
Measurement Gap and Decision
The Measurement Gap
Traditional inspection management relies on compliance specialists to read inspection reports and classify findings manually. This works when inspection volume is low and specialists are experienced. AI-powered classification promises to handle higher volumes and provide consistent classification, but without FCAR, compliance teams cannot verify whether the AI's classifications match what the regulatory authority intended.
The gap is particularly dangerous because NLP classification errors are silent. A human misclassifying a finding creates a conversation: someone might question it. An AI misclassification enters the system with the same confidence as a correct one, and downstream remediation workflows execute against it without review.
The Decision It Supports
FCAR tells compliance leadership whether the AI classification system is trustworthy for routing inspection findings to remediation workflows. An FCAR above 90% with under-classification below 5% means the system is reliable and conservative. An FCAR below 80% or an under-classification rate above 10% means the system is misrouting findings at a rate that creates compliance risk. The metric enables specific action: retrain the NLP on recent reports with expert-labeled classifications for any jurisdiction or finding type where accuracy is low.
Formula
Plain language: Of the findings that an expert reviewer determines are violations (code deficiencies requiring corrective action), what fraction did the AI classify as observations? That under-classification rate is the primary metric. Overall classification accuracy is secondary.
Primary metric: Violation Under-Classification Rate (False-Negative Rate):
FCAR_FNR = Under_classified / (Correctly_classified_violations + Under_classified)
Secondary metric: Overall Classification Accuracy:
FCAR_OA = Correctly_classified / Sampled_findings × 100
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Under_classified | Violations classified as observations by the AI (dangerous direction: false negatives) | Count |
| Correctly_classified_violations | Violations the AI correctly classified as violations (true positives) | Count |
| Correctly_classified | All findings where AI classification matched expert reviewer | Count |
| Over_classified | Observations classified as violations by the AI (conservative direction) | Count |
| Sampled_findings | Total findings reviewed by the expert | Count |
Why false-negative rate is primary: In facilities management, overclassification (observations classified as violations) creates unnecessary corrective action tracking. Under-classification (violations classified as observations) means fire code deficiencies, life safety violations, or OSHA-reportable conditions are documented as minor observations and may never receive corrective action. The asymmetry is a consequence structure, not a preference. A 90% overall accuracy that under-classifies 10% of code violations is a life safety liability, not acceptable performance.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Is the NLP correctly interpreting inspection reports, or misclassifying violations as observations?