Well, the future of heuristic contests looks pretty bleak (as we expected). This contest was pay to win.
I went through all of the source codes of the top20 contestants and it looks that all of them (except one or two) feature a very heavy usage of agents (e.g. bloated code, tons of arbitrary magic constants, lots of nearly dead code that handle esoteric cases that you would never ever cover yourself).
To be fair, this was quite unusual contest that relied on a standard idea and the whole difficulty was about implementing it and covering all of the possible edge cases. It's hard to imagine a better problem for an AI agent.
