# AI quality monitoring worksheet
Use fictional or approved information. Example counts are not operational thresholds.

## Measurement contract
- Task and permitted use:
- Time window:
- Population and sample selection:
- Number reviewed / total eligible:
- Rubric and configuration versions:
- Pending or unreviewable cases:

## Separate indicators
| Indicator | Numerator | Denominator | Slice | Owner | Trigger | Response |
|---|---|---|---|---|---|---|
| Service health | | | | | | |
| Source-supported answers | | | | | | |
| Exception handling | | | | | | |
| Input mix | | | | | | |

## Worked calculation
- A: routine 76/80; exceptions 14/20. Total:
- B: routine 38/40; exceptions 42/60. Total:
- Group rates:
- What changed:
- What these samples cannot establish:

## Investigation and recovery
- Observation:
- Hypothesis:
- Evidence that could disprove it:
- Named response owner:
- Temporary restriction:
- Recovery checks and retest date:
- Logging access and retention:

Source guide: https://learn.aicompetence.org/knowledge/monitor-ai-system-quality/