Relay / fictional worked example
Verification: evidence matrix
The aim is to tie checks to claims. A green build is not a general declaration of correctness.
| Claim | Evidence to gather | Limitation |
|---|---|---|
| Identical external replay does not create another request | Contract test of second submit | Demonstration is in memory, not a concurrent distributed system |
| Conflicting duplicate is not silently merged | Contract test for changed payload | Doesn’t prove reconciliation with a real upstream source |
| Invalid input does not create state | Validation and store-size test | Covers specific validation rules only |
| Creation audit is emitted once | Audit test across first delivery and replay | Demo does not prove atomic durability with persistence |
| Assignment requires proper authorisation | Permission and tenant integration tests | Not implemented in reference code |
| Assignment resists concurrent overwrite | Version conflict tests | Not implemented in reference code |
| UI supports the product’s ownership workflow | Moderated user tasks, observation, follow-up | No real user evidence in this illustrative case |
| Market position is differentiated | Competitor research and buyer interviews | Not conducted; proposition is a hypothesis |
| Release is safe | Security review, data protection, reliability tests, rollout/rollback | Not provided by teaching example |
Independent checks versus generated explanations
An implementing agent may produce a completion statement, but it is the test runner, source diff, contract artefacts and independent reviews that provide assessable evidence.
A review can find a missing test even after existing tests pass. A passing test can support a behavioural claim even if two models argue about implementation style. Neither eliminates the need to validate product value with people.
Quality bar for the intake slice
- No schema or contract changes hidden inside implementation.
- Unit tests pass for creation, replay, conflict, validation, audit behaviour and independent submissions.
- No external dependencies or additional persistence mechanism introduced.
- Any gaps in durability, concurrency and tenancy are explicitly called out as gaps.
To view the actual test output for the packaged executable slice, see ../../04-executable-example/TEST-RESULTS.md. That record should only describe checks actually run.