Purpose & scope

• Intended use, exclusions, target users, operational constraints, safety context, sector overlays.

Model card template

• Summary; inputs/outputs; data sources; training recipe; evaluation suite; safety & fairness; limitations; changelog.

Datasets & cohorts

• Representativeness; cohort definitions; exclusions; labelling quality; leakage checks; licensing and lawful basis.

Metrics & thresholds

• Per-cohort accuracy; parity metrics; robustness indices; uncertainty calibration; thresholds and waiver justifications.

Robustness/adversarial

• Stress, distribution shift, adversarial prompts; mitigations; incident learnings fed back into tests.

Explainability & limits

• Appropriate explanation modes; uncertainty cues; disclaimers; human-in-the-loop guidance.

Evaluation runbook

• Data splits; repeatability; environment capture; seeds; tooling; reviewers; sign-off path.

Acceptance & sign-off

• Acceptance report; exceptions/waivers; residual risk rationale; go/no-go decision and owners.

Post-release validation

• Shadow evaluation; cohort regression; PMM metrics; triggers for rollback/model updates.

Versioning & lineage

• Link model card versions to training data and evaluation runs; immutable snapshots per release.

Evidence pack links

• Direct URLs to evaluations, dashboards, risk register, approvals and incident bundles.

Evaluation checklist

• Template complete; thresholds per cohort; runbook repeatable; approvals signed; links live.

© Zen AI Governance UK Ltd • Regulatory Knowledge • v1 05 Nov 2025 • This page is general guidance, not legal advice.