How granular is your test tagging strategy and does it actually get used?

We use pytest marks for test organisation: `@pytest.mark.slow`, `@pytest.mark.integration`, `@pytest.mark.smoke`. In theory: great. In practice: - Devs forget to add marks - Marks drift from reality (a test marked `smoke` that now takes 2 minutes) - We have 6 marks and nobody can remember which tests belong to which mark Do you use a tagging/marking strategy that's actually maintained? Is the answer fewer marks (2–3 maximum) or a different enforcement mechanism?