Starcraft AI blog

Funny, I see it the other way. I find win% round robin to be a silly format, so all my tournament-specific preparation is for the elimination bracket.

Congrats on another playoff appearance, and good luck!

Please don't accuse me of bias without evidence. Especially when we've discussed the topic at length before on multiple occasions and you know I have a well-considered case, even if you disagree with it. It's one thing to disagree with my arguments and a whole other to accuse me of bias.

I've made the case against win% round robin formats before and will continue to do so whenever given the platform. I don't think they make sense as a format because:

1. Human StarCraft tournaments are not played with the expectation of winning nearly 100% of games. They're played with the requirement of winning best-of-N series. Having to prepare for win% round robin renders our bots less prepared for play against humans.
2. The requirement of winning ~100% of games eliminates large swaths of the human metagame. For example, 12 Hatch in ZvZ becomes totally unviable against opponents who are going to 4pool more than ~5% of the time. This leads to conservative, repetitive play, and less interesting games.
3. The requirement of winning ~100% of games assumes that the game is balanced across race matchups. It's not. Trying to win TvT 100% of the time is way easier than trying to win ZvZ 100% of the time, because TvT is rarely won or lost in a single moment, but ZvZ often is. I think the format overall is really bad for Zerg.
4. The format encourages shoring up tiny weaknesses against weaker bots rather than improving overall play against stronger bots. This produces games that are both less fun to work on and less interesting to watch.

You've asked me before what I think the alternative is. I think round robin qualifiers for an elimination bracket are good (with the elimination bracket being the focus). But to create a better event that's pure round robin, I think the formula is this:
1. Bots are ranked on how many opponents they had winning matchups against
2. First tiebreaker: Head-to-head record against all tied opponents
3. Second tiebreaker: Overall win%

This format encourages authors to produce stronger play in the human dimension (eg. trying to win series against strong opponents, not trying to stomp weaker opponents with perfect efficiency). It'd lead to more interesting games, better bot-vs-human play, and frankly would be a lot more fun to work on as an author.

There's a lot to the topic! Would love to hear your take.

Dan, your alternative ranking sounds interesting, but I would add another ranking parameter:
Add 0.5 for each 1-1 matchups.
Just like in chess, a winning matchup is 1, a draw matchup is 0.5 and a loss matchup is 0. This year Microwave has a lot of 1-1 matchups, and I think that 1-1 should still count for something.

Ohh right! :D
Well... 1-1 matchups needs to count somehow... I can't se they should not.