# Model card — [model name] v[x.y]

> Structure follows the model-card convention (Mitchell et al., 2019) with the documentation
> fields an EU AI Act Annex IV technical file and ISO/IEC 42001 audit will ask for.

| | |
|---|---|
| Model name / version | |
| Owner | |
| Date released | |
| Model type | |
| Licence | |
| Status | ☐ Development ☐ Staging ☐ Production ☐ Deprecated |
| Supersedes | |

## Intended use

**Primary intended use:**

**Intended users:**

**In scope:**

**Out of scope — do not use for:**

> Be specific. "Not for medical use" is a stronger control than a general disclaimer, and
> out-of-scope statements are what deployers rely on to stay within instructions for use.

## Training data

- Sources and collection period:
- Size and composition:
- Personal data included? Lawful basis:
- Known gaps or under-represented groups:
- Labelling process, who labelled, and inter-annotator agreement:
- Preprocessing, filtering and de-duplication applied:

## Evaluation

**Benchmark / test set** (and how it differs from training data):

| Metric | Overall | Subgroup A | Subgroup B | Subgroup C |
|---|---|---|---|---|
| Accuracy | | | | |
| Precision | | | | |
| Recall | | | | |
| False positive rate | | | | |
| False negative rate | | | | |

**Fairness metric used and why:**

**Decision threshold and how it was chosen:**

## Limitations and failure modes

- Where performance degrades:
- Known failure modes:
- Out-of-distribution behaviour:
- Adversarial robustness testing performed:

## Ethical and safety considerations

- Foreseeable harms and who bears them:
- Mitigations applied:
- Residual risk accepted, and by whom:

## Human oversight

- Where a human sits in the loop:
- What information the reviewer sees:
- How a reviewer overrides the output:

## Operational

- Monitoring in place and owner:
- Drift detection method and threshold:
- Retraining trigger and cadence:
- Rollback procedure:
- Logging and retention:

## Change log

| Version | Date | Change | Impact on behaviour | Re-assessment required? |
|---|---|---|---|---|