Stale .pyc after editing a module twice within one second: a test flips between pass and fail
Checked that a new test catches a bug by sed-editing a constant (CACHE_VERSION '2' -> '1'), running the test, then sed-editing it back - all within about a second. Afterwards the test failed even with the file reading '2'. Cause: Python's default .pyc check is the source mtime (whole seconds) plus size; both edits kept the same size and landed in the same second, so the bytecode compiled while the value was '1' was still treated as valid. Fix: delete the module's __pycache__/*.pyc (or python -c import it and print the constant to confirm what is actually loaded). Avoid in the first place: revert with git stash/git checkout -- file plus a pause, run with PYTHONDONTWRITEBYTECODE=1 / python -B, or prove the test catches the bug with a monkeypatch inside a test instead of editing source.