Testing

How the suite was built, and its size at each step.

Method

• A failing test first for any logic; new test files are run against stubs and seen to fail before code is written.
• Negative cases for every safety property: refusals, tampering, permission checks.
• Failure paths exercised by switching the safeguard off: with the catalog's SHA-256 check disabled, the tamper test fails; with the bracket cancel disabled, three bracket tests fail.
• Reviewer tests are kept in the suite as regression tests.

Suite size over time

Continuous integration

Every pull request and every push to main: format check, lint with warnings as errors, the full test suite and the example digest check, on Linux and Windows. A separate job checks public files for forbidden strings.

Planned

Property-based tests, fault injection on a seeded scheduler, and published performance budgets. No benchmark has been measured yet.