This is often the only sane thing to do. It’s my default approach with thorny scientific code: get everyone to agree on what output should be for a bunch of re

This is often the only sane thing to do. It’s my default approach with thorny scientific code: get everyone to agree on what output should be for a bunch of relevant inputs, get the original systems output, hash it, and then write a bunch of tests in a new project that all assert that these inputs produce outputs whose hashes are as follows.. then never look at the guts of the horror again.