§0statement
§1why
Why it works
A deterministicGives the same result each time, with no language model. check gives the same answer every run, so a failure is a fact. A model's judgment varies, so it can rank and suggest but should not gate.
§2example
Copy this
# CI order
- run: tz check && tz lint # deterministic
- run: npm run check:raw-values # deterministic
- run: npm run review:voice # model, advisory: comments, never fails the job
§3breaks
Where it breaks
Some rules have no deterministic form, such as tone. Those stay advisory and are labelled so.
§4cases
Used in: Case 4, Case 5, and the voice-and-tone plugin.