A green suite is a claim.
This is where it gets checked. Write tests, and the engine breaks the code underneath them — systematically, one edit at a time — and counts how many of those breaks your tests notice.
Coverage says a line ran
It cannot say whether anything looked at the result. A suite can execute every branch of a module and assert nothing about any of them, and the badge will read 100%.
This counts what your tests caught
The engine changes the code one small edit at a time — the kind of edit a person or a model actually makes by mistake — and runs your suite against each broken version.
Nothing here is a judgement call
Your score is arithmetic over executions that actually happened. There is no rubric, no reference suite to match, and no model deciding whether your tests are any good.
The demonstration
A suite with 100% line coverage and a mutation score of zero.
Every line executes. Not one assertion would catch a change to any of them. It takes about ten minutes, and afterwards a coverage badge never quite reads the same way again.
Take the lesson