pin-the-case-that-narrowed-the-rule

Trigger

A check you wrote has failed on something that is actually fine, and you are about to add an
exemption, an allowlist entry, or a narrower pattern.

It does not apply when the check found a real problem — fix the problem. It also does not apply to
tuning a threshold before the check has ever run in anger.

Do

1. Write down, in one sentence, why the flagged case is legitimate. If you cannot, the check may be
   right and the code wrong.
2. Add both cases to the test suite: the legitimate one that must pass, and a synthetic version
   of the real violation that must still fail.
3. Narrow the rule as little as possible. Exempt a path, not a whole directory; require a negation
   nearby, rather than deleting the phrase from the list.
4. Put the narrowing in its own commit, and say in the message which case forced it.

Avoid

Deleting the rule because it was annoying once. Also avoid the quieter version: broadening an
exemption until the rule covers nothing — an allowlist that grows every sprint is a rule being
retired one line at a time, without anyone deciding to retire it.

Avoid exempting by file when the real distinction is by meaning. A phrase check that bans a
sentence outright will ban the warning that negates it, and the honest fix is to require the
negation, not to stop checking that page.

Check

After narrowing, run the suite with the original violation reintroduced. It must still fail.

Then read the diff of the rule itself and ask: what class of problem can now get through that could
not before? If you cannot answer, the narrowing was not understood.

Risk

This adds a test for every exception, and a suite full of exception tests is a suite that is
tedious to read.

The bigger risk is treating the pinned test as proof the rule is still strong. It proves one case
still fails. A rule narrowed five times has five pinned cases and possibly a large hole between
them, and only re-reading the rule as a whole will show that.