test-neu

A section inside an existing test file is the rule; a new file is the exception. The check may join only if it shares the same claim, the same ground, and the same name truthfulness. One no means a new file, which then owes a row in the tests table written by overview.py.

Use the pattern <subject>_<claim>_test.py, at most 24 characters before _test.py, lowercase English, starting with one of twelve fixed prefixes like files_, sound_, or table_. The prefix says where the fault would sit, and the second half is a claim, not a thing.

A bare assert throws a traceback instead of a readable line, stops at the first failure, carries no numbers, and is not counted. Every verdict goes through the canonical check function, which prints a named line, collects failures, and feeds the closing summary.

A counter-proof shows the check goes red when the thing it guards is actually broken, using a deliberately broken copy of the program. Each check needs its own proof recorded in tests/state/counterproof, keyed by the check's first-argument wording.

Only while the test is being written, never in the finished version. Waiting must be on a condition with a short interval and a generous upper bound, measuring standstill rather than elapsed time, and exhausted patience must end red with a line saying what never came.

Common causes are assuming the folder equals the repository, relying on gitignored working notes, or missing environment variables that run.sh sets like LANG, TMPDIR, and VPM_FIXTURES. Verify with a git archive clone including git init and git add -A, and always run through run.sh.