Duplicate chips != skill
Six mirrored tables, 600 hands, 30k-hand oracle: duplicate winnings tied raw chips. Two and three tables were worse. Shared deals do not cancel interaction.
PokerOn
A few hands anyone can spike a river. PokerOn does not crown that player. Every action is scored against a frozen model. Runout luck is stripped to zero. After a short session the leaderboard is who played better - the unbiased game of skill.
Six mirrored tables, 600 hands, 30k-hand oracle: duplicate winnings tied raw chips. Two and three tables were worse. Shared deals do not cancel interaction.
After lock, luck = award - equity x pot. Mean luck over remaining runouts is zero. Rank uses regret vs Q, not the spike.
600 decisions: 0.15 BB random Q error kept Spearman 0.983 with latent skill. 0.50 BB persistent bias collapsed it to 0.631. Wrong Q is worse than noisy Q.
A checked flop does not cap every range in a 3-way pot. We abstain on that class instead of scoring it with heads-up Q. See the method note.
Information I never includes the future board or the payout. Rank cannot be rescued by a lucky river.
Locked all-ins keep an expected award. Variance is shown, then peeled off. Everyone’s luck meter aims at 0.
Weighted Elo on every fold, check, call, raise. Cluster-weighted so one street cannot spam the rating.
Play, see regret vs the best action, climb the same progress bar. Practice never contaminates unaided Elo.
What you are actually climbing
Chip graphs are sports entertainment. StackRank is the other column: Skill Elo, mean regret, and a shared bar to publication. Identities, simulations, and the abstention rule live in the method note. The house still holds the shuffle.
Continue with Google, or email. After signup you land on your desk - play, stats, the common ladder.
Continue with GoogleI already have emailStackRank Labs - Unrated · certified GTO Hold’em Elo is not claimed. This is the skill channel the table uses.