Skip to content

Adapter Matrix

The adapter matrix is the evidence ledger behind user-facing coverage. Each adapter wraps one public-service channel or mockable shape as one tool entry. Without this ledger, the docs would be only claims.

Users do not need to read adapter IDs first, but evaluators and contributors do. A Live statement should trace back to an adapter or generated metadata entry that explains primitive, state, permission, schema, and citation.

The generated adapter data currently represents three broad groups:

  • live find adapters for public lookup domains such as weather, road, hospital, emergency, and welfare guidance;
  • location and administrative-area adapters that support locate;
  • mock check or send adapters for identity, certificate, authentication, MyData, protected submission, or payment-shaped workflows.

This split mirrors UMMAYA’s trust model. Public lookup can often be Live earlier. Protected completion requires stronger authority and usually remains Mock or Handoff until official access exists.

A useful adapter is not only a function. It must carry enough metadata for the query engine, permission layer, docs, and evaluator to agree.

FieldWhy it matters
tool IDstable reference for docs, traces, and generated metadata
primitivetells the model whether the path is locate, find, check, or send
tierdistinguishes Live, Mock, Handoff, or Planned state
permission tierprevents protected work from becoming silent execution
schema pathvalidates arguments and output shape
citation or sourceproves that the adapter follows an external boundary

If a field is missing, the adapter may still be code, but it is not ready to support a strong documentation claim.

The matrix protects users from vague coverage language. When a page says UMMAYA can find public safety information, the adapter evidence should show which public lookup path supports that claim. When a page says a payment flow is Mock, the matrix should prevent the final answer from sounding like a paid bill.

That is why adapter metadata is part of user trust, not only developer inventory.

The canonical adapter catalog lives in docs/api/README.md. Generated metadata is copied into docs-site/src/data/generated/adapters.json and /_llm/generated/adapters.json.

After adapter changes, run:

Terminal window
npm run docs:generate
npm run docs:check

If generated metadata changes but prose does not, review the affected pages before publishing. That is the docs drift gate.