One approach that's probably quite easy to implement is to generate test suite code coverage by file, and only do the substitutions in files with sufficient cov

One approach that's probably quite easy to implement is to generate test suite code coverage by file, and only do the substitutions in files with sufficient coverage (where your definition of "sufficient" might vary, of course).