What makes information hidden
In many card games, you know your own cards and whatever has been played face up, but not what the other players hold. That gap is hidden information. You cannot remove it, but you can reason about it: some holdings are more likely than others given what you can see, and what you can see changes as the game goes on.
Two kinds of evidence are involved. The first is about the cards themselves: which cards are visible, and so which cannot be in anyone’s hidden hand. The second is about behaviour: what a player chose to do, and what that suggests about their cards. Both can be handled with the same updating rule, Bayes theorem. The treatment here is general and educational. It is not a method for winning, and real games involve far more uncertainty than the toy examples below.
Evidence from the cards you can see
This part is pure counting and is the more reliable of the two.
Take a standard 52-card deck. Before anything is dealt, the chance that a given card is an ace is 4/52, about 7.7%. Now suppose you are dealt two aces and can see no other cards. From your point of view, 50 cards are unseen, and 2 of them are aces. The chance that one particular unseen card, say the first card in an opponent’s hand, is an ace is 2/50 = 4%.
So your own cards shifted the chance from about 7.7% to 4%. You learned nothing about the opponent directly. You only learned that two of the aces are accounted for. The same logic applies to every card that is revealed: each one removes a possibility from the unseen pool and changes the counts. This is the dependence at work in dependent events, and the arithmetic for combinations of unseen cards is the territory of drawing specific cards and the hypergeometric distribution.
Evidence from what players do
Behaviour is trickier, because it depends on how players choose their actions, which you do not know. To see the mechanics, take a deliberately simple illustration.
Suppose an opponent is one of two kinds of holding at some point: “strong” or “weak”. Say that, before any action, you think 20% of the time they would be strong. Further suppose, purely for the sake of argument, that a strong holding leads to a raise 70% of the time and a weak holding leads to a raise 10% of the time. The opponent raises. How should that change your view?
Use 1,000 imagined occasions:
- Strong: 20% of 1,000 = 200. A raise happens 70% of the time: 140.
- Weak: 800. A raise happens 10% of the time: 80.
Raises in total: 140 + 80 = 220. Of those, 140 come from strong holdings, so the updated chance of strong is 140 / 220 = 7/11, about 64%.
A raise moved the estimate from 20% to about 64%. It is not a certainty, because weak holdings sometimes raise too. If the opponent raised weak holdings 30% of the time instead, there would be 240 raises from the weak group, and the updated chance of strong would be 140 / 380, about 37%. The conclusion depends heavily on the assumed behaviour, which is the uncertain part.
Why the inputs are the weak point
The arithmetic is exact, but the inputs are guesses. Some things to keep in mind:
- Players adapt. If an opponent suspects you are reading their actions, they may deliberately change them. The assumed rates can then be wrong in a way that is hard to detect.
- Situation matters. The same action can mean different things at different points of a game, with different stack sizes, scores or positions.
- Small samples mislead. A few observations of a player’s habits are weak support for a firm claim about their behaviour, a point developed in the discussion of small sample sizes.
- Ranges beat single hands. Rather than guessing one exact holding, it is more sensible to think in terms of a spread of possibilities, which is how hand ranges are described in poker discussions.
Updating step by step
Information arrives in pieces, and the update can be done in the same order. Start with a prior from the cards alone. After one action, update. After the next action, take the result as the new prior and update again. A tree with branches for each holding and each action keeps this tidy, and counts per 1,000 occasions are easier to follow than a chain of percentages, as explained in the piece on natural frequencies.
Two cautions apply when chaining updates. First, the later actions must be treated as separate evidence only if they are not just repeats of the same signal. Second, a long chain accumulates the error in each guessed input, so the final number is often much less precise than its decimal places suggest.
Limits and honest expectations
Reasoning about hidden information can sharpen understanding of a game. It cannot guarantee any result. Cards are shuffled randomly, the next card is not influenced by earlier plays, and a good read can still lose to an unlucky deal. In games played for money, the reasoning above does not remove the house edge in banked casino games, and even in player-versus-player games, outcomes are uncertain and skill comparisons only matter over a long run of results.
Gambling can lead to real harm. Never bet more than you can afford to lose, never chase losses, and if you feel your play is getting out of hand, reach out to a local support service.
The takeaway
The method is the same one used everywhere in Bayesian reasoning: begin with what you know, ask how likely each possibility would be to produce what you saw, then renormalise. With cards, the first step is firm counting. With behaviour, the second step is a guess, and the honest conclusion is a range of plausible answers rather than a single confident number.