Back to Guides
A short AI-narrated film with English captions. Pressing play loads the YouTube player. Watch on YouTube
Guide Origin: Official | ArcKit Version: [VERSION]
/arckit:atrs generates an Algorithmic Transparency Recording Standard (ATRS) record for AI and algorithmic tools used in public services.
Inputs
| Artefact | Purpose |
|---|---|
Requirements (ARC-<id>-REQ-v1.0.md) |
System purpose and user impact |
| AI Playbook assessment | Responsible AI context |
| Data model | Data sources and processing |
| Stakeholder drivers | Business context and affected groups |
Command
/arckit:atrs Create ATRS record for <algorithmic tool>
Output: projects/<id>/ARC-<id>-ATRS-v1.0.md
ATRS Record Structure
| Section | Contents |
|---|---|
| Basic Information | Name, owner, description, deployment date |
| Purpose and Scope | Why the tool exists and what decisions it supports |
| How It Works | Technical overview (accessible to public) |
| Data Used | Input data, sources, personal data handling |
| Human Oversight | How humans are involved in the process |
| Risk Management | Identified risks and mitigations |
| Impact Assessment | Who is affected and how |
| Performance Monitoring | How accuracy and fairness are measured |
| Contact Information | Who to contact for questions |
ATRS Tiers
| Tier | Description | Disclosure Level |
|---|---|---|
| Tier 1 | Low impact, simple algorithms | Basic disclosure |
| Tier 2 | Moderate impact, ML-based | Standard disclosure |
| Tier 3 | High impact, complex AI | Full disclosure |
One-Page Workflow
| Phase | Key Activities | ArcKit Commands |
|---|---|---|
| Discovery | Define algorithmic tool and impact | /arckit:requirements, /arckit:stakeholders |
| Assessment | Assess AI compliance and risks | /arckit:ai-playbook, /arckit:dpia |
| Transparency | Create ATRS record | /arckit:atrs |
| Publication | Publish record to gov.uk | Manual process |
| Review | Regular review and updates | Annual cycle |
Review Checklist
- Tool purpose and scope clearly described in plain language.
- Technical description accessible to non-technical readers.
- All data sources identified with personal data flagged.
- Human oversight mechanisms documented.
- Risk mitigations in place for identified risks.
- Impact assessment covers all affected groups.
- Performance metrics defined for ongoing monitoring.
- Contact information current and accessible.
Publication Requirements
| Requirement | Description |
|---|---|
| Format | Structured format per CDDO guidance |
| Location | Published on gov.uk algorithmic transparency hub |
| Review Cycle | Annual review minimum |
| Updates | Material changes require record update |
Key Principles
- Plain Language: Records must be understandable by the public.
- Proactive Disclosure: Publish before being asked.
- Living Document: Update when the system changes.
- Honest Assessment: Acknowledge limitations and risks.
- Accessibility: Information available to all affected citizens.