Two Leopards in Five Zha Jin Hua Hands: Is the Deck Rigged?

A player wrote in: in five hands of Zha Jin Hua, the table saw two Leopards (three of a kind), so is something wrong with the deck? We checked the code and the hand records and found no sign of manipulated dealing or an unusually high Leopard rate. Every Zha Jin Hua hand uses a full 52-card deck shuffled by a random number generator seeded from the operating system, and each player then receives three cards in order; the deal does not look at who is playing or who is winning, and nothing adjusts hand categories. Bots read only their own cards after looking, and judge opponents only from public bets and comparison results. As of 29 October 2026, all 11,397 Zha Jin Hua hands dealt 59,192 three-card holdings, 158 of them Leopards, or 0.27%; the theoretical rate is 52/22,100, about 0.24%, and the gap is within normal variation. Counting only the 19,034 holdings dealt to human players, there were 45 Leopards against an expected 44.8. Every Leopard in the player’s screenshots matches the records card for card; the two that went to the same player were 62 hands and more than half an hour apart, not within five hands. Across all of the reporting player’s hands, the number of Leopards at the table matched the expected count.

So why can Leopards feel common? One player holds a Leopard 1 time in 425, but at a six-player table someone has one in about 1.4% of hands, and two in five consecutive hands happens about 0.2% of the time; with one to two thousand hands a day across the platform, someone will see it. A Leopard that wins alone is also shown to the table and collects a bonus from every player, so it sticks in memory while the ordinary hands nobody shows are quickly forgotten. Each Zha Jin Hua hand now publishes a deal commitment before play; afterwards you can verify that hand’s deal from the table’s hand history, or download the proof and check it with the public offline verifier. As long as the browser that saw the hand start still holds the commitment it saved (up to 30 days), the proof detects a deck swapped afterwards, but the server still knows the cards during the hand, and hands dealt before this update have no proof (see fairness scope). The findings on Leopard frequency above come from code review and full-history statistics.