RunOnProofDocs
Evidence firstScoped decisions
Integration · Relevance AI

Fixtures and expected results

The seven fixtures test the decision contract without a live verification or payment. Company numbers beginning with ZZ are deliberately synthetic and must never be described as Companies House evidence.

Run the repository suite with:

npm run demo:relevance-ai
npm run test:relevance-ai

Run one scenario with:

node integrations/relevance-ai/demo.mjs company-active

1. Company active

  • Fixture: company-active
  • Expected Tool: Company Check
  • Expected recommendation: PROCEED
  • Expected economics: no quote settlement, charge, or production mutation
  • Important assertion: evidence, freshness, coverage, and limitations remain visible

This fixture proves mapping behavior, not the live status of a real company.

2. Ambiguous company

  • Fixture: ambiguous-company
  • Expected recommendation: REVIEW
  • Expected next action: request the exact legal company number
  • Forbidden behavior: selecting one legal entity by guesswork

3. Source unavailable

  • Fixture: source-unavailable
  • Expected recommendation: REVIEW
  • Expected economics: no decision and no charge
  • Forbidden behavior: translating source failure into no finding or approval

4. Conditioned capability

  • Fixture: conditioned-capability
  • Expected recommendation: STOP
  • Expected network activity: none
  • Expected explanation: the required qualified external evidence or certified source is unavailable

5. Payment required

  • Fixture: payment-required
  • Expected recommendation: REVIEW
  • Expected product: gb.solution.company_check.v1
  • Expected price: 0.04 USDC
  • Expected behavior: require explicit external payment authorization; redact payment material
  • Expected economics: no payment and no settlement

6. Replay

  • Fixture: replay
  • Expected recommendation: PROCEED for the fixture result
  • Expected flag: replayed=true
  • Expected source activity: zero new live-source calls
  • Expected product charge: 0.00 USDC additional

Replay proves retrieval of a committed execution, not current source freshness beyond the stored validity window.

7. Idempotency

  • Fixture: idempotency
  • Identical payload and key: converge on the original request/result
  • Changed payload with the same key: HTTP 409 conflict
  • Forbidden behavior: inventing a new key to conceal the conflict

Workspace QA order

  1. Run repository tests first.
  2. Test the Tool’s normalization with copied synthetic response bodies rather than a production POST.
  3. Test the agent’s explanation and Tool selection.
  4. Confirm Requires approval remains enabled.
  5. Inspect history for zero unintended production calls and payments.

Evidence to record

Record test date, repository SHA, Tool version, agent version, cases run, pass/fail, and reviewer. Record economic delta separately. Do not save secrets, payment material, or real personal data.

The complete seven-case contract is certified by this repository's synthetic tests. Do not describe those fixtures as live company checks or as proof of a private Relevance AI workspace execution.

RunOnProof public documentationExplicit coverage · scoped claims · evidence first