A mimetic is a kind of test which mimics the implementation of the code rather than testing the behavior. They are a common form of unit test.
How do you know if you've got a mimetic test?
A fully mimetic test fails when 100% of the failures of the test are caused by code being changed rather than the presence of a bug or lack of desired functionality.
Note that flakiness is different to mimeticism.
What if a test fails 10% of the time solely because code changed?
Congratulations, your test is useful! In practice almost every test will be at least slightly mimetic, so a small number of failures caused by code changes is ok, provided this is acknowledged as a bug (however minor).
What's wrong with mimetic tests
Mimetic tests not only impose a cost in terms of the cost to build, they also impose an ongoing maintenance cost as changes to the code will require changes to tests.
Mimetic tests also don't catch much beyond the most trivial bugs (e.g. syntax error).
A high proportion of mimetic tests often leads to test abandonment.
Why does anybody write mimetic tests if they are so useless?
Mimetic tests tend to crop up when developers write code first and then decide afterwards that their code needs some tests. They see classes and think "I should write some tests for that class or method". Sometimes these are goodhart's tests.
Classes are implementation details. Methods are implementation details.
Unit tests lend themselves naturally to mimeticism simply because they are inherently designed in such a way as to test implementation rather than behavior.