Yeah, I noticed that when I put the chart in up above. I’m not sure if I’m super concerned about it, though. It encourages people to shy away from 7-3 matchups (or 3-7 matchups) unless they’re pretty confident that they can bend the percentages up (assuming all players are rational and/or I do lots of explaining up front), which should lead to us seeing less-frequently-played matchups played more often (which is one of my goals for the tournament).
I would think that would be lessened by the differing rates of point accumulation. In regular Swiss each round can only ever get you 1 pt. In this format, you could high-roll your way into a ton of points after making peanuts in a few rounds. Like, every round you have a chance of “winning more than 1 set” in standard Swiss.
Ah, this is the much bigger issue. If you have byes of any kind, you’re going to need to figure out some way to pay out the player who got the bye (or won against a no-show opponent). Maybe you do average points earned by all non-bye players for that round, instead of a flat value? I could see a flat value either being garbage, in which case you’d never want the bye, or too good, and then you’d prefer the bye to playing at all.
Yeah, I like the “average of points earned by other players” option. Or maybe median, because median is usually better than average for almost all things.
Alternatively, you could rank players by the mean or median of their own personal net scores for every set they did play. That way it doesn’t matter how many sets they actually play?
Yeah, using the mean could work. Using the median would have the effect of filtering out matches where you did really well, which isn’t an effect I would want, I think.
Although, maybe that would lead to pathological behavior where someone won really big in the first round, and could then just forfeit out of the future matches to keep the average? I guess keeping the IYL 2-forfeits-and-you’re-out rule would avoid that.
Ok, I think I’ve talked around this idea enough that I’m just going to go ahead and run it as a tournament, and see who signs up and how it goes. One final tweak I was thinking about is that I’d raise the number of games in a match to 5, which should more closely line up with a Bo7 experience as far as match time goes.
The main thing I’d watch for is players picking into matchups that they don’t consider well represented by the historical data. Like, a matchup that was 2-8 historically due to low sample size, but in reality is much closer.
Honestly, that’s an outcome I’d be happy to see. If this tournament pushes people into unexpected counterpicks and new tech, that would be a win in my book.
As I was computing the historical matchup chart to put into the tournament intro post, I actually ran the numbers to see how confident we can be in particular matchups in that historical chart. In this chart, the number in each box is the 95%ile confidence interval for the historical matchup number (I’m not a good enough statistician to correctly define with the confidence interval is, but roughly, if the confidence interval is small, that means we can be more sure that the matchup number is accurate).
As you can see, there’s actually a pretty wide margin on most of these matchups, because of the small sample size, but in particular, BBB/Ven, BBB/Menelker, BBB/Jaina, BBB/Gloria and Gloria/Ven are especially uncertain. (Looking at @mysticjuicer’s records, they’ve all only been played about 30-40 times).