System overview & purpose

• Intended purpose, exclusions, users, environments; architecture diagrams; interfaces and dependencies.
• Functional decomposition; decision scopes; human oversight pattern and control points.

Datasets & lineage

• Training/eval/hold-out descriptions; provenance, licenses, quality and bias assessments.
• Lineage graphs; transformations; sampling; anonymisation/pseudonymisation.

Models & artifacts

• Model cards; hyperparameters; features/embeddings; prompt templates for LLMs; RAG index sources.
• Artifact hashes; SBOM; attestation; trusted build chain; vulnerability posture.

Metrics & thresholds

• Performance targets by cohort; fairness metrics; safety constraints; explainability acceptance.
• Alert thresholds; owners; dashboards; escalation mapping.

Security & robustness

• Threat model; adversarial testing; content filters; rate-limits; isolation; secrets management.
• Backups; disaster recovery; dependency hardening; SBOM/SCA policies.

Explainability & UX

• Global and local methods; rationale displays; warnings; confidence and uncertainty hints.
• Accessibility; language support; adverse case UX flows.

Deployment & operations

• CI/CD pipeline with policy gates; approvals; environment parity; observability schema.
• Runbooks; on-call strategy; incident collaboration; ticket workflows.

Updates, rollback & change control

• Controlled updates with evaluation diffs; dark-launch; feature flags; staged rollout; rollback criteria.
• Change advisory approvals; communication plan; user re-training notices if behaviour materially changes.

User instructions & limitations

• Clear instructions per role; safe prompts/patterns; limitation statements; prohibited uses and escalation.

Evidence & approvals

• Link to risk file, DPIAs, security tests, bias audits, acceptance reports, governance minutes.

Annex: schemas & templates

• Model card template; data card; lineage schema; incident bundle structure; CAPA template.

Implementation checklist

• All sections complete and versioned; links to evidence live; approvals recorded.
• Auto-exports from MLOps enabled; document ownership assigned.

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