Testing And Evidence¶
Goal¶
Tie claims to evidence. This repository should not rely on confidence language when current tests or maintainer checks can answer the question directly.
flowchart TD
A[Code or doc change] --> B[Targeted tests]
B --> C[Workspace checks]
C --> D[Maintainer checks when needed]
D --> E[Reviewable evidence] This diagram shows the evidence ladder the repository expects. Targeted tests come first, broader checks follow, and maintainer-only checks are added when a surface needs them before review can be considered complete.
stateDiagram-v2
[*] --> Changed
Changed --> Tested
Tested --> Checked
Checked --> Reviewed
Reviewed --> Merged The state diagram expresses that same rule as a workflow boundary. A change is not ready to merge simply because code exists; it has to move through testing, checking, and reviewable evidence.
Evidence Rules¶
- do not describe work as done if the relevant tests are red or missing
- generated artifacts are disposable outputs, not repository truth
- prefer live tests and maintainer commands over hand-maintained status prose
- if a public behavior changed, update the tests and docs that describe it
Claim Discipline¶
- treat missing evidence as an open gap, not an implicit success
- review
cargo run -q -p bijux-dev-cli --bin bijux-dev-cli -- status --format json --no-prettybefore making current maintainer status claims - if parity, docs, or diagnostics still report
partial,missing, or equivalent blockers, say that directly instead of rounding it up to done
Stability Review¶
Treat a surface as stable enough to describe as current only when:
- route behavior and command identity are covered by current tests or contract checks
- stdout, stderr, and exit-code behavior are covered where the surface is part of the public CLI
- install, plugin, or state diagnostics do not report unresolved blockers for that area
Treat a surface as still risky when:
- parity rows are
partialormissing - lifecycle write paths still rely on unproven recovery behavior
- docs or tests still describe the area as intentionally different or unresolved
Common Verification Commands¶
cargo test --locked --workspace
python3 -m pytest crates/bijux-cli-python/tests/python
cargo run -q -p bijux-dev-cli --bin bijux-dev-cli -- status --format json --no-pretty
cargo run -q -p bijux-dev-cli --bin bijux-dev-cli -- parity --format json --no-pretty
cargo run -q -p bijux-dev-cli --bin bijux-dev-cli -- docs-audit --format json --no-pretty
make docs-check
Review Rule¶
A convincing change explanation should answer:
- what changed
- what evidence proves it
- what still remains uncertain, if anything
For release or review summaries, keep the outcome shape explicit:
doneleftblockedordeferred
Honest Limit¶
Passing tests do not make every architectural claim true by themselves. Tests and docs have to agree, and unsupported claims should still be stated as unsupported.
Read Next¶
Continue to Release And Compatibility.