@planet @clojure How do you guard against the case where you generate incorrect output because of a bug, but fail to notice because spotting errors in generated output is harder (albeit less work) than constructing correct output manually? For me it’s a big issue, but no doubt varies a lot across people and domains. My only solution is “git good”, i.e. try harder.

@planet @clojure strictly, these are characterisation tests rather than unit tests. Characterisation tests are tests to check that the behaviour of a program doesn’t change; unit tests should be tests that the program is working as specified.