Mutation testing can be a ratchet. When new code is introduced, the numbers must still meet thresholds. That tells us resulting changes do not leave the codebase worse off than before.

Threshold numbers are stored alongside the code. For pitest (Java), those are test coverage, mutation coverage, and test strength.

Baseline is per-repo, so comparisons are always to the repo itself. Existing projects start at when thresholds are introduced. New projects start at a higher bar.