Interpret tests and delivery receipts
A receipt proves the commands and observations it records for a named scenario, target and version. It does not become a current runbook or establish every related workflow. Read the boundary of the result before relying on its pass/fail label.
Source review, test execution and runtime observation are different evidence. Reading a test specifies what it is intended to exercise; running it establishes the outcome on its actual inputs. Observing a public endpoint establishes only that endpoint’s returned fact at that time.
Receipt reference
Section titled “Receipt reference”| Evidence item | Record or inspect | Interpretation limit |
|---|---|---|
| Source review | Full revision, owning files and declared question | Describes implementation; no execution implied |
| Local check | Command, synthetic/local target, input revision and outcome | Mocks/fixtures do not prove provider behaviour |
| CI run | Workflow, selected jobs, actual checkout and result | Combined merge tree may differ from branch artifact |
| Runtime observation | Component, environment, time and returned identity/status | No automatic schema or business acceptance |
| Business scenario | Actor/resource, action, expected/actual outcome and version | Other actors, records or receiving systems remain separate |
| Recovery drill | Exact generation, component integrity, identity/startup and timing | Component-only proof cannot become complete recovery |
| HTML distribution | Clean source commit, manifest, ZIP hash and extracted-browser result | No hosted publication or combined PDF implied |
Execution and reuse
Section titled “Execution and reuse”Reuse established results when relevant inputs are demonstrably identical. Compare the code, migration, fixture, configuration contract and check implementation that determine the result. State the previous executed revision and the scope of the comparison. A new documentation paragraph does not require repeating unchanged provider actions or a full application suite.
Do not reuse a result when its target, source, policy, fixture or material dependency changed. Run the focused affected check and expand only when a new failure or dependency justifies it. CI may independently select broader suites; let the actual selected run finish rather than dispatching a duplicate workflow.
A failed check needs its failed step and candidate identity preserved. An unchanged transient infrastructure failure can justify retrying failed jobs. A deterministic error needs correction. Do not relabel a skip, cancellation or absent selected job as success.
Build and writing evidence
Section titled “Build and writing evidence”The handbook’s build-info.json identifies the documentation source commit, baseline and worktree state. Page metadata retains each article’s own reviewed source. A clean build commit can differ from the implementation revision an older article reviewed; those are intentional separate identities.
Static verification checks every registered page/topic address, source reference, contextual relationship and local asset. Browser checks exercise reader flows, local diagrams/search, mobile matrices and print without external runtime requests. Packaging records archive membership and hashes. Independent extraction and serving checks the actual distributed files. None of these checks exercises a live invoice or establishes independent business acceptance.
Historical receipts and current instructions
Section titled “Historical receipts and current instructions”Keep dated evidence immutable. Use docs/MAP.md and active runbooks for the current method. If a later method fixes a gap, add its scoped evidence without rewriting what the older drill proved. A database and Storage receipt from one generation remains component evidence until a combined application/identity check exists.
Do not copy private payloads or secrets into receipts to make them appear detailed. Use sanitized outcomes and restricted evidence locations. Logs and credentials own access and exposure boundaries.
Synthetic interpretation example
Section titled “Synthetic interpretation example”A synthetic test can prove that a refused cross-company request performs no business mutation and that an allowed request succeeds. A receipt that records only the allowed request cannot establish the negative boundary. A mocked provider receipt cannot establish real receiving-system acceptance.
Likewise, a green archive browser check establishes the distributable handbook, while an API /readyz check establishes its named runtime prerequisites. Report each result beside its limitation. Status, test levels and recovery define the next evidence needed when a broader claim matters.