Testing

Each case under tests/fixtures/scan/ has an answer key written by hand that grades every usage: migrated, or left and why. The first five were written before the scanner existed. The suite reproduces every key, and each rewritten file matches its expected file byte for byte. tests/fixtures/scan/COVERAGE.md lists 38 gaps the fixtures do not cover, 20 of them closed. CI runs the suite on Linux for every supported Python and on macOS and Windows for the oldest, the newest and 3.14, and holds each system to 100% branch coverage of the code it runs.

The benchmark measures the same code on public repositories, and CONTRIBUTING.md says how to run the suite.