INDIAN POKER

Full Rules

The deck and the deal

A full 54-card deck — 2 through Ace in four suits, plus a monochrome joker and a colour joker — is shuffled and split into two private 27-card stacks, one per player. Each round both players draw one card from their own stack and place it face out on their forehead without looking. Ranking runs 2 up to Ace, then the monochrome joker, then the colour joker at the top; within a rank the suits break the tie in the order clubs, hearts, diamonds, spades. Because every card is unique, a showdown can never be tied. Over 27 rounds every card in the deck is used exactly once, so a player who counts carefully knows more with every round that passes.

Antes and betting

Each player starts with 2,000 chips. In rounds 1 to 22 the player who acts second posts an ante of 20. From round 23 the ante climbs every round — 80, 130, 210, 340, and 550 in round 27, where both players post it. Betting then runs over three fixed-limit streets at 20, 20 and 40 chips per bet, with at most four bets or raises in a street. Fold at any point and your opponent takes the pot immediately — and your card is swept away unseen, which matters more than it sounds. If a player cannot cover the ante they post what they have, and an empty stack ends the game on the spot with the pot going to the survivor; otherwise whoever holds more chips after round 27 takes it.

The rising ante

The last five rounds used to be decorative. Whoever led at round 23 won 81% of the time, and a lead of 1,000 or more converted 98.6% of the time. The first fix was an all-in the trailing player could declare, and it failed for a reason worth stating: a leader is never obliged to accept it. Chips have collapsing marginal value once you are ahead — a player up 1,000 needs about 72% equity before calling an all-in is correct, where pot odds would ask for 48% — and that threshold rises the further ahead they get. Widening the window did not help; it only bought the leader more free folds. So the escape hatch was replaced with something nobody can decline. The ante itself climbs: 80 at round 23, then 130, 210, 340, and 550 from both players in round 27, which opens that pot at 1,100 before a single bet. Measured over 1,200 games per setting, this pulls a round-23 lead of 1,200 or more back from 95% to 81%, and a lead under 300 down to 51%, while only 7% of games end early on an empty stack. The endgame is live again, and it is live for both players at once.

The three givers

After the first betting street, each player privately picks two of the three givers and asks each one whether their own card beats the opponent's. One giver tells the truth 80% of the time, one 50%, one 20% — the assignment is fixed for the whole game and never revealed. You only find out whether a giver lied to you if the round reaches showdown, because only then do you learn what your card actually was. Fold and those two answers stay unverifiable forever. Your opponent never learns which givers you asked or what they told you, so the two of you build completely different pictures of who is trustworthy. Counting cards feeds directly into this: when an extremely low card is on your opponent's forehead you already know the truthful answer, so you can grade a giver for free.

Buying information

After the second street you may spend chips on two kinds of information. The Monty Hall costs 100, then 200, then 400 for the first, second and third use in a game: it narrows your own card down to two candidates, the real one and a decoy, so you know one of them is on your forehead but not which. The lookback costs 50 and doubles each time you use it, and reveals up to three of the cards you have already burned in earlier rounds without ever seeing — which is how you convert a fold-heavy early game back into a sharp count. Chips spent this way are burned rather than added to the pot, so a game can end with fewer chips on the table than it started with, and all three purchases close down when the blind phase begins in round 23. Since round 27 reveals your opponent's last card and therefore pins down exactly which 27 cards were yours, the lookback is worth most precisely when it is no longer for sale — buy it early or not at all.

The handover

Between rounds 10 and 15 a third purchase is on the table, and only one of the two of you can ever make it. For 300 chips the three givers hand their roles to successors you arranged: the 80/50/20 assignment is reshuffled into a different one, you are told the new assignment exactly, and your opponent is left with nothing — every inference they built over the first ten rounds now points at people who are no longer in the chair. Answers already given that round are void for both of you. The purchase is announced, so they know they have been reset, but not what to. It is priced against a measurement: ten rounds is enough for most players to pin down the 80% and the 20% giver, and the whole giver system is worth roughly 300 chips across a full game. A blind reshuffle that erases both sides symmetrically is worth about 90 chips and is a trap at any real price; the version where the buyer knows the new arrangement is worth about 170, which puts break-even near 240. At 300 it is a real decision rather than a free win — the answer to a game in which the reading has already been done.

Why bluffing runs backwards here

In ordinary poker you bet to misrepresent your own hand. Here your opponent can already see your hand, so a bet says nothing about what they hold and everything about what you hold. Measured over thousands of simulated games: when the opponent opens the first street, your card averages 17 on a 0–53 scale; when they check, it averages 40. One check is worth twenty-two points of information about a card you cannot see. That is why random bluffing loses money in this game — throwing a bet in with a weak read does not confuse your opponent about your hand, it simply hands them a false reading of theirs at a price you pay. The skill is symmetric and it is the whole game: read their actions, and do not be readable yourself.

The four across the table

The Rookie does not count cards and believes every giver is honest. The Player counts perfectly and tracks which giver is which by Bayesian inference, but throws your betting away as noise. The Sharper reads your bets as evidence about its own card — which is what the structure of this game actually rewards, and it beats the Player 84% of the time on that one idea alone. Laplace held that probability is only the measure of our ignorance, and that an intellect knowing the state of everything would have no need of it. The demon named after him starts from the Sharper's reading and then tunes it to you specifically: every showdown reveals its own card, which retroactively tells it exactly what you were looking at when you bet, so its read of you sharpens with every hand it survives to the end. Fold a lot and you starve it of exactly that data — the same structure that governs your own verification of the givers, pointed back at you. Its calibration error against a fixed opponent falls from 0.169 to 0.097 over a thousand games, and it does not win by pushing harder, it wins by being right more often.