Oh well. I don't think I would have won anyway. The problem seems to be that the REPL behaves very differently to non-REPL contexts, which I certainly wasn't expecting. Actually, it looks like eval in the REPL uses the REPL namespace as a default (so for example you can type (eval '(define a 2)) (write a) and see a 2 printed out) which seems rather unconscionable to me¹, especially since the REPL namespace includes a boatload of miscellaneous modules such as racket/pretty and racket/function and probably others as well.

Possibly next time it would be useful to publish a "test harness" ahead of time which runs contestants against each other in a clean (or at least well-specified) namespace, and make it clear that the actual contest would be run using this code.

¹ Actually, now that I've noticed this, it looks like you could use this to write a bot that corrupts the global (REPL) namespace, and therefore the namespace of anyone trying to simulate it, maybe even forcing the opponent to cooperate by rewriting if or something. Ha!