So a while back some people asked me about the predictive accuracy of the Aligulac system. They didn't feel this beautiful graph was enough, and maybe rightly so. The rating system has been optimized over the whole database, while people are generally only interested in the high-level games. These are harder to predict and the picture looks a bit different if you restrict your view to only the top.
So I accepted the challenge and started playing Liquibet. Now, I don't really know how the seasons work there and I have no idea when it will be reset, so I just started in the middle. That's why I'm not currently leading.
I once said that I thought a good liquibetter would beat Aligulac. Well, turns out maybe the difference isn't as high as I thought it was.
So far I have won 46/72 points just blindly following Aligulac predictions. For 1v1 that means inputting player names here and then just picking the highest number. For groups, that means generally picking the two with the highest probability of advancing, though there are some caveats to this since the choices aren't independent. For team matches it means picking whichever team has the highest rank in the relevant format. Some teams are not listed (such as EG-TL), so in those cases some minor mental effort is required.
46/72 points is a rate of 63.88%. The current leader is SpiZe, who has 161/253. That is 63.64%.
Does this mean Liquibet is broken?
A game that offered higher payouts for predicting upsets (like every other betting game in existence) would be both harder to break systematically and probably also more exciting to play.