In the past code bullet tournament, I noticed something interesting in the tournament. My algo got beaten quite quickly in the 4th round
Masteroliw(Highest elo: 1806)'s algo ‘pizza_time_algo_V0.2.12’ beat kkroep(Highest elo:2000)'s algo ‘EMP^ERROR_v1.0’ in 33 turns - https://terminal.c1games.com/watch/627743
The main reason for losing this battle is due to an experiment in this algo that tries to anticipate large scale attacks and counter with scramblers. I found that this particular idea is effective against more algorithms than not, so I left it in. However against the very specific strategy where the opponent doesn’t attack at all and saves up very big, it fails spectacularly (see replay).
It is an interesting example of why I think the current competition method is flawed. I think algorithms should be encouraged to win most of the matches all algorithms, and preferably all their matches against the top dogs. There are a lot of trade-offs to consider to counter different strategies. Countering all requires an extremely perceptive algorithm.
Having the entire tournament be single-elimination increases the risk of algorithms that attempt adaptability be eliminated by accidental hard counters.
I think the solution would be to make the tournament style multi-stage, with round robin groups at the start. This way only top dogs with superior strategies and/or adaptability end up in the single elimination tournament.
What do you guys think?
PS: I wouldn’t have won the tournament regardless, as I’m pretty sure both finalist comfortably beat my algo