Coolers at Loose Tables

The short answer

Some players have told us they see more Texas Hold'em coolers here than elsewhere and suspect the dealing. Others think the cause is simply that people play too many hands. We re-audited the open-source dealing code and ran simulations on the same shuffle and hand-ranking code. Two findings:

• We found no dealing bug that would make coolers more common. The card order is fixed by a random seed before any betting, and cards are then taken from fixed positions.
• The "everyone plays too many hands" explanation holds up. In the model below, raising every player's VPIP at a 6-max table from about 15% to about 40% raises the chance of a cooler from 0.89% to 4.17% of hands, about 4.7 times.

In this article a cooler is a showdown where the loser holds at least three of a kind made with their hole cards: either a higher hand category than the five board cards alone, or a bigger straight, flush or full house when the board already shows one. Set over set and a flush losing to a bigger flush are examples. Beating the board only on a kicker does not count.

What the dealing code does

1. Seed. Ordinary Texas tables, including games against bots, are dealt by a separate sealed dealer process. It draws a 32-byte server seed from the operating system's random source and publishes a commitment first; players whose client supports the protocol then contribute their own random seeds. The server seed, the player seeds received and the hand id go through SHA-256 to give deck_seed, which becomes the 256-bit ChaCha20 key. Tournament tables skip this process and use 32 bytes from the operating system's random source directly as the same key.
2. Shuffle. Deck::new runs a Fisher–Yates shuffle over 52 cards, using rejection sampling to avoid modulo bias.
3. Deal. Hole cards go around the table twice: the i-th dealt seat gets cards i and i+n. The five board cards are positions 2n to 2n+4. There are no burn cards.

The seed is the only input to the shuffle. Who entered the pot, how much they bet, their past results and their membership play no part, and no card can be swapped on the turn or river. We found no live Texas dealing path that selects a preset deck; test hooks that override hole cards or pin a seed compile only in test builds and are absent from production builds. Short Deck and the other games use their own decks and shuffles.

We also tested the production shuffle statistically. We split 12 million decks into 120 groups and computed a position-by-card chi-square for each. The theoretical mean is 2652 and the observed mean was 2650.5; the 120 p-values showed no significant departure from uniform (KS test D = 0.045, 5% critical value 0.124). Separately, in one million decks two given seats both received a pocket pair 0.3500% of the time, against an exact value of 0.3505%.

Why loose play creates coolers

A cooler needs at least two players to reach showdown and both to make a big hand. The more players enter and the wider their ranges, the more often both happen at once.

The table below models a 6-max table where everyone plays only the strongest part of their starting hands (ranked by estimated equity against one random hand and cut at whole hand classes, so the actual VPIP is slightly below target: 14.8%, 21.4%, 29.4%, 39.5%, 59.3% and 100%) and goes to the river once in. Each row is one million hands:

The effect is stronger at nine seats: 1.96% of hands at about 15% VPIP, 8.37% at about 40% and 14.48% at about 60%.

The last column is closest to how it feels at the table. For each hand you play, a table where everyone plays about 15% hands you a cooler loss about once every 95 hands; at about 40% it is about once every 50. Wide ranges hold more small pairs, flush draws and middle connectors, and when those connect they run into your strong hand.

The model assumes nobody folds after entering. Real players fold after the flop, so the absolute numbers in the table are too high. What the model shows is the direction: with everything else equal, more players entering with wider ranges means more coolers. It is not a prediction of real table frequencies.

Why it feels like so many

A cooler is usually the biggest pot of the session, and the player who lost it remembers it; nobody remembers a quiet fold. If players at a table generally enter more loosely, or bomb pots are on (everyone posts the same amount and goes straight to the flop), more players see the flop and coolers become more common. That can explain why it feels like more, but we have not checked it against real hands on this platform.

If you doubt a particular hand, open Fairness detail from the table menu after it ends and let your browser recompute the deck. You can also download the proof and recompute it offline with the open-source verifier; see Verify a hand yourself. These checks show that the card order matches the pre-deal commitment; they cannot prove that the operator never cheats. For those limits, see Cooler Hands and Dealing.

This article uses simulation only and reads no real player hands. Code entry points: engine/src/rng.rs, engine/src/deck.rs, server/src/session/hand_play.rs, server/src/sealed_dealer.rs, server/src/pf_dealing.rs.