Skip to main content
lab-notes.aiPublic-service AI playbooks

Contribute / improve the starting point

Make the research pack more useful.

Five useful places to contribute.

Choose the layer you can improve. The linked Traffic Flow Managementfiles are concrete examples; use the equivalent path for the playbook you are changing.

Final check / npm run check

  1. 01

    Opportunity copy

    Make the strategy opportunity clearer, more bounded and easier for a builder to understand. Keep product choices open and claims no stronger than the evidence.

    Open an example playbook definition (opens in a new tab)npm run test -- content/playbooks/content.test.ts
  2. 02

    Source verification

    Check that a registered source still resolves, that its access label is accurate and that the stated coverage and relevance match what it publishes.

    Inspect the registered source format (opens in a new tab)npm run test -- content/playbooks/content.test.ts
  3. 03

    Synthetic working data

    Improve a safe stand-in using only structures and vocabulary supported by published sources. Preserve the disclosure and state what the file cannot prove.

    Inspect an example dataset (opens in a new tab)npm run test -- content/playbooks/content.test.ts
  4. 04

    Domain brief

    Strengthen vocabulary, stakeholder context, source boundaries, known unknowns and the questions a builder should answer before choosing an approach.

  5. 05

    Build-partner instructions

    Improve how the checked-in skill distinguishes fact, interpretation and synthetic data; explores unranked directions; and stops when outside authority is needed.

Non-negotiable

Privacy comes first.

Do not commit names, contact details, identifiers, exact addresses or real person-level health, justice, education, housing, benefits or consultation records. Dataset tests walk every committed value against lib/privacy-patterns.ts.

If a responsible stand-in cannot be made, improve the playbook’s refusal and state what authorised work would need instead.