- Error messages about output being different from the expected use some sort of very explicit printing mode with lots of quotes and escaping. A message saying that one got "\\\'\\\'" where "\'\\\\'" was expected is needlessly hard to parse when the actual difference is \'\' vs. '\\' (or whatever).
- Many parts are marked optional, but the test script complains if you don't implement them, and later sections assume that they are implemented, so you have to treat them as compulsory anyway.
- Some things regarding environment updates were contradictory or at least not explained properly at first, with some very relevant information only coming later. I don't remember the specifics, but the problem was roughly that I set up my data structures thinking that everything could be implemented in a pure way, but at some point it became clear that you either need some other data structure or must use destructive modifications.