Test Strategy¶
bijux-gnss-dev needs layered proof because it governs repository behavior
rather than one isolated algorithm.
Proof Route¶
flowchart TD
files["governed inputs"]
commands["maintainer commands"]
tests["integration tests"]
benchmark["benchmark comparison"]
reviewer["reviewer confidence"]
files --> commands --> reviewer
files --> tests --> reviewer
benchmark --> reviewer
Main Proof Layers¶
| layer | protects | first command or test |
|---|---|---|
| command execution | command-line behavior over governed inputs | audit-allowlist, deny-policy-deviations, audit-ignore-args |
| repository guardrails | crate scope, paths, and workflow boundaries | integration_guardrails.rs |
| suite selection | slow-test roster integrity and nextest selection behavior | integration_nextest_suite_selection.rs |
| governed-input consistency | documented files match command and test expectations | docs plus source review |
| benchmark comparison | baseline extraction and regression reporting | bench-compare |
Layer Discipline¶
- A passing command does not prove the repository boundary is still sound.
- A passing guardrail test does not prove a governed-file workflow still means the same thing.
- A benchmark comparison can report insufficient evidence when a full benchmark run is too expensive for the current documentation pass.
- New governed files need both command or test proof and reader-facing docs.
- Failures should name the repository contract that was violated, not just the file path that failed parsing.
First Proof Check¶
Inspect crates/bijux-gnss-dev/docs/TESTS.md,
crates/bijux-gnss-dev/tests/integration_guardrails.rs,
crates/bijux-gnss-dev/tests/integration_nextest_suite_selection.rs,
crates/bijux-gnss-dev/docs/GOVERNANCE_FILES.md, and
crates/bijux-gnss-dev/docs/BENCHMARKS.md.