KEY TAKEAWAYS
CONTENTS
Online poker looks like a table on a screen, but the table exists on a server. When a hand starts, software on that server shuffles a digital deck, deals each player’s cards, checks that every bet is legal and affordable, decides who wins and moves the money. The app on a phone or computer shows the hand and passes the player’s choices back to the server. It does not decide anything about the cards.
That design is not only a convenience. A technical standard for interactive gaming systems published by Gaming Laboratories International, GLI-19 version 3.0, says player software must not contain logic that generates the result of any game, and that critical functions, including generating any game outcome, belong to the gaming platform and stay independent of the player’s device. In a game where players compete against each other, keeping the cards on the server also helps keep one player’s device from learning another’s hand.
This guide explains the parts of the software, how the shuffle and the deal work in principle, what happens when a connection drops, how tournaments differ from cash tables, and how game integrity is protected in general terms.
How an Online Poker Hand Works: Client, Server and Table
Two programs cooperate. The client is the app or browser page the player sees. The server is the platform software that hosts tables, holds account balances and runs the game. GLI-19 describes the first as player software and the second as the gaming platform or interactive gaming system, and expects the platform, not the device, to generate outcomes.
A table is a software object on the server with a list of seats, a deck, a pot and a record of whose turn it is. The server moves it through a fixed sequence for every hand: seat and blinds, shuffle, deal, betting rounds, showdown, payout. Each step has rules the server enforces on every action, so a client that sends an impossible request, such as betting more than the balance or acting out of turn, is simply refused.
Poker also differs from house-banked games in one respect. GLI-19 notes that in peer-to-peer sessions the operator usually does not take part as a party; it provides the environment and may take a rake, commission or fee. The software therefore acts as a neutral dealer and bookkeeper between players rather than as an opponent.
Finally, the platform records everything. Game history lets a hand be reconstructed afterward, which matters for disputes and reviews and is the basis for the checks described later.
Under the Hood: Shuffling, Randomness and Dealing
The shuffle is where randomness enters. A standard deck has 52 cards, so the number of possible orderings is 52 factorial:
where 52! is the number of distinct orderings of a 52-card deck. A random number generator (RNG) must be able to reach every one of them. A generator that starts from a 32-bit seed has only 2^32 = 4,294,967,296 starting states, so it could produce at most that many different shuffles, a vanishingly small fraction (about 5 × 10^−59) of the possible orderings. Counting bits, 2^225 is smaller than 52! while 2^226 is larger, so a generator needs at least 226 bits of internal state to be able to produce every deck.
The standard reflects this. GLI-19 version 3.0 requires that the set of possible outcomes be large enough for all outcomes to be available on every draw, that scaling, mapping and shuffling algorithms be unbiased, and that the RNG be cryptographically strong and not seeded from a time value alone. An independent test laboratory reviews the source code for shuffling algorithms and statistically tests the final output, after shuffling, at a 99% confidence level. The wider RNG story is covered in GamingTechNow’s guide to casino RNGs.
Once the deck order is fixed, dealing is bookkeeping: the server assigns cards to seats in order and reveals community cards only when the rules call for them. The deal is logged so that game recall can reconstruct what happened, and GLI-19 requires that the final outcome of each game stay visible long enough for a player to verify it. The laboratory review covers the generator, the scaling or mapping and the shuffle as separate pieces, so a problem such as a biased mapping is examined even when the generator itself is sound.
Walkthrough: One Hypothetical Hand at a Six-Seat Table
This example is hypothetical and simplified; real implementations vary. Suppose six players are seated at a hold’em table.
- Seating and blinds: the server confirms each player is seated once and moves the blind amounts from account balances into the pot.
- Shuffle: the server uses its RNG to produce a fresh ordering of the 52 cards.
- Deal: two cards go to each of the six seats, 12 cards in all. In a typical design, each client receives only its own player’s cards.
- Action: when it is a player’s turn, the client sends a choice. The server checks it against the rules and the balance, then updates the pot.
- Community cards: the server reveals the flop, turn and river in turn, up to five shared cards, so at most 17 of the 52 cards are dealt to players or the board.
- Showdown and payout: the server ranks the hands, awards the pot, deducts any rake and writes the balances.
- Record: the hand is stored so it can be shown in game recall or reviewed later.
Variations: Cash Tables, Tournaments and Interrupted Hands
The same engine supports different formats. The table compares general features, noting that product details vary by supplier and state.
| Feature | Cash-game table | Tournament |
|---|---|---|
| Money | Players buy in with funds from their account balance | Players pay an entry fee and receive tournament chips |
| Records | Hand history and balances | Entry fees, player rankings, prizes paid and event status, which GLI-19 lists among tournament data |
| Typical extra software | Seating and waiting lists | Blind schedules and moving players between tables as others are eliminated |
Connections also fail. GLI-19 version 3.0 treats a hand as interrupted when communications are lost, the platform or device restarts, or the software ends abnormally. Wagers stay held by the platform until the hand completes, and in multi-player games, if a player cannot act in the allotted time, the platform completes the hand on that player’s behalf under the game rules and shows which decisions it made. One player not completing an action must not change how other players are credited or debited.
A related feature is the Away from Play status, which skips a player’s turns after a request or a period of inactivity and, after a disclosed time, removes the player from the session.
Myths, Limits and Misconceptions
Myth: the software knows who needs a card. GLI-19 requires that the game not modify or discard RNG outcomes through adaptive behavior, and that events of chance be independent of earlier games. The standard prohibits steering a shuffle toward or away from any player.
Myth: a slow connection changes the cards. The standard says events of chance must not be affected by bandwidth, link utilization or error rate. Connection quality can affect whether an action arrives in time, which is why players must be told where speed or disconnection could matter.
Myth: a certified shuffle guarantees every result feels random. Streaks and bad beats are normal in any random process, and testing checks distribution and independence, not how a session feels.
Limits apply as well. Standards set minimums, not designs, and the exact methods used differ by supplier, state and software version. Nothing in a public standard shows how any particular site is built.
Key Facts and Integrity at a Glance
The essentials, including the protections built around the game:
| Topic | What to know |
|---|---|
| Who decides outcomes | The server, independent of the player’s device |
| Shuffle | An RNG reviewed and statistically tested by an independent laboratory |
| Seating | Players should be prevented from occupying more than one position in a session |
| Monitoring | Operators are expected to detect collusion and unauthorized player software |
| Reporting | Players must be able to report suspected cheating |
These protections come from standards such as GLI-19 and from regulators, such as the Alcohol and Gaming Commission of Ontario, whose standards expect operators to deter, prevent and detect collusion and cheating and to log related activity. Specific methods are deliberately not published in detail, and rules vary by state. A player who suspects cheating should use the site’s reporting channel.
Related Concepts and Next Steps
To go deeper on the randomness behind the shuffle, read how casino RNGs work. For the platform that hosts tables, accounts and wallets, see how an online gambling platform works behind the screen. For the same RNG principles inside a physical cabinet, read how slot machines work.
Natural follow-ups include how tournament software balances tables as players are eliminated, how game history and hand records are stored and reviewed, and how detection systems recognize unusual play in general terms. Readers interested in the testing side can also follow how independent laboratories certify a shuffling algorithm and what a certification does and does not show.
Frequently Asked Questions
