# AI Output Validation Checklist — The Facility AI Institute

Run these checks before an AI-generated number or recommendation leaves your desk.

## Data
- [ ] Was the input deduplicated (e.g., by work order ID)?
- [ ] Were category labels normalized (e.g., "hvac" vs "HVAC")?
- [ ] Were missing or invalid values (blank dates, negative durations) handled and disclosed?
- [ ] Were outliers checked before averaging or ranking?

## Calculations
- [ ] Did I recalculate at least one headline figure myself?
- [ ] Is the denominator correct and stated?
- [ ] Do totals reconcile with the source system?

## Claims
- [ ] Is every causal claim ("X reduced Y") supported by before/after evidence?
- [ ] Does any recommendation rest on a single record or outlier?
- [ ] Are assumptions written down?

## Governance
- [ ] Has a named person reviewed and signed off?
- [ ] Is it clear which parts were AI-generated?