Dragon Tiger Predictor: Can Anything Call the Next Round?

Every method below is a rule a predictor might follow. Each one sees only the decided results of earlier rounds in the current shoe, makes its call before the deal, and starts every shoe with nothing to go on. Ties are counted but not scored, because no one called a tie. The last column bets one unit on every call under the rules you choose, so you can see what the hit rate is worth at the table.

How each method makes its call

• Always Dragon: Calls Dragon every round.
• Always Tiger: Calls Tiger every round.
• Follow the last winner: Calls whichever side won the last decided round.
• Bet against the last winner: Calls the side that lost the last decided round.
• Back the hot side: Calls the side with more wins in the last ten decided rounds.
• Back the side that is due: Calls the side with fewer wins in the last ten decided rounds.
• Pattern matcher: Finds every earlier place in this shoe where the last three results appeared, and calls whichever side followed them more often.
• Coin toss: Calls a side at random.

We ran the eight methods over 16,000,000 rounds from eight-deck shoes on October 13, 2026 (1,194,037 rounds tied, 14,805,963 decided). One standard error of a hit rate at that size is 0.013 points, so anything inside about 49.97% to 50.03% is indistinguishable from a coin. The result per unit bets one unit on every call.

The exact cost of any Dragon or Tiger bet is 3.73% of the stake with half lost on a tie and 7.47% with all of it lost, whoever or whatever chose the side. The measured column wanders a few hundredths either side of those figures, which is ordinary sampling error. To break even, a method would have to be right on 52.02% of decided rounds under the half-lost rule and 54.04% under the all-lost rule. The best of eight managed 50.01%.

The shoe is shuffled

A round of Dragon Tiger is the next two cards off a shoe of several decks, usually eight, shuffled before play and dealt until the cut card. The order was fixed at the shuffle. Nothing that happens at the table, no streak, no road, no bet, moves a card. A predictor that looks at past results is reading the same shuffled order as everyone else, from the wrong end.

Dragon and Tiger are exactly even, always

This is the part that makes Dragon Tiger different from baccarat, where drawing rules give Banker a small, known advantage. Here the two cards come from the same shoe, and the shoe does not know which one is dealt first. Whatever cards remain, the chance that the first card outranks the second equals the chance that the second outranks the first. Dragon and Tiger are even on round one of a fresh shoe and even on the last round before the cut, with five Dragons in a row behind you or fifty. That is a statement about the cards, not a measurement, and the measurement above agrees with it to the second decimal.

The one thing the cards do change is the tie

Past cards do carry information, but about the wrong thing. When a rank runs out of a shoe, a tie becomes a little less likely; when ranks bunch up, more likely. Counting every card in 16,000,000 rounds of eight-deck play, the chance of a tie on the next round ran from 6.08% to 12.65% against the fresh-shoe 7.47%. The Tie bet at 11 to 1 was favorable on 0.511% of rounds, about one in 200, by 3.71% of the bet on average. A player who counted every card and bet the Tie only then would earn 0.0189% of a unit per round dealt: a chip every five thousand rounds or so, before the casino noticed. Dragon against Tiger did not move at all.

What real players do with the scoreboard

A 2025 study in Acta Psychologica analyzed 17,970,830 games of baccarat, the sister game, played by 6,625 customers at a casino in South Korea between April 2017 and August 2019. After a single Player win, customers bet Player 46.6% of the time; with every further win in the streak the odds of betting with it rose by about 38%, and Banker streaks drew the same response. More than nine customers in ten showed the pattern, and bets grew by 2% to 3% per step when following a streak. Once a streak was broken, the pattern vanished. The authors call the trends coincidental, with no rational basis for betting on them, and note that following one does not change the expected return. Dragon Tiger's roads are the same scoreboard with one card a side, and our own figure for following the last winner is 50.00% right, the same as betting against it.

Some predictors claim to call the card itself: high or low, red or black, odd or even. On a fresh eight-deck shoe each rank is 32 of 416 cards, 1 in 13 (7.69%); a card of 8 or higher is 6 ranks in 13 (46.15%), as is a card of 6 or lower, and the 7 that loses both Big and Small bets is the other 7.69%. Red and black are exactly half each. Those are the figures a casino prices its one-card side bets on, and they are why the one-card bets that lose on a 7 carry the 7.69% edge shown on our Odds & Payouts page. The only honest way to move them is to count the cards that have gone, which shifts the Big and Small bets by a little and Dragon against Tiger by nothing. Our Winning Strategies page works through counting the 7s and finds the gain too small to use.

Search for a Dragon Tiger predictor online and the results are app-store listings, APK downloads, “AI prediction software” sold by subscription, pattern charts in PDF form, and videos promising a strategy with a high win rate. Under the different names they do one of three things.

• They keep score. Most predictor apps are trackers: you type in each result and the app draws the bead road, big road and the other scoreboards, counts streaks and alternations, and points to the next cell a pattern would fill. That is a record of what has happened, the same record the casino shows you for free. The more careful app listings say as much in their own descriptions: the predictions are statistical estimates, not a winning method and not a guarantee.
• They sell a percentage. “AI” and “machine learning” predictor sites quote accuracy figures and win rates. On a standard eight-deck table Dragon wins 46.27% of all rounds, Tiger wins 46.27% and the rest are ties, and every method we tested, over 16,000,000 rounds, called the winner of the rounds that were not ties right between 49.98% and 50.01% of the time. A win rate of 77%, a figure one strategy video puts in its title, would mean doing what a fair coin does when it gets 77 of 100 calls right, which happens about once in 36 million tries. A seller who could do that over a real stretch of play would not need to sell anything.
• They ask for something. APK files from outside the official app stores, “mod” versions, paid “prediction codes” and subscriptions are where the risk sits: an app installed from an unknown source can carry malware or ask for permissions it has no use for, and a subscription takes money before the first round is dealt. Nothing they offer can change the cards.

This page is the opposite kind of predictor: it runs in your browser with nothing to install, no sign-up and nothing to pay, and it shows you every call it gets wrong. Try the method tester at 100,000 rounds and watch every method settle at a coin toss.

Short runs are the predictor's best friend. A fair coin gets at least 7 of 10 calls right 17.2% of the time, and at least 60 of 100 right 2.8% of the time, about once in 35 tries. A seller who shows you the one run in 35 has shown you nothing. Over 1,000 calls a coin lands between about 46.9% and 53.1% nineteen times in twenty; over 100,000 the band is a few tenths of a point. That is why the tester above lets you run 100,000 rounds, and why our own figures come from 16,000,000.

Two other tricks are worth knowing. A tool that reports only its wins, or restarts its count after a loss, will show a high rate forever. And a claim of "90% accuracy" about a game with two sides is a claim to have found something no mathematician has, because a method that was right even 55% of the time would earn 5.52% of every unit bet under the half-lost rule, and no casino could keep the game on its floor.

Try it: run the tester at 100 rounds a few times. One method or another will show 58% or 62%, and it will be a different method each time.

Apart from not working, a predictor can be a crime to carry. Nevada's NRS 465.075 makes it unlawful to use, or possess with intent to use, any device, software or hardware designed to obtain an advantage at playing a game in a licensed gaming establishment, and it names devices that project the outcome, keep track of cards, analyze probabilities or analyze strategy. Under NRS 465.088 a first offense is a category C felony with any gain disgorged to the state; a second is a category B felony carrying one to six years. The statute does not care whether the device works. Other jurisdictions have their own laws and casinos have their own rules. We are not lawyers; this is what the Nevada statute says, read on October 12, 2026.

Nothing predicts the round, but three things tell you exactly what it costs. Our Dragon Tiger Probability Calculator gives the exact probabilities and house edges for any deck count, tie payout and tie rule. The Dragon Tiger Simulator deals real shoes with pretend chips, and its betting system tester shows that staking systems do to the cost per chip what predictors do to the hit rate: nothing. And our Winning Strategies page says what is left once prediction is off the table: the lowest-edge bet, a slower table, and a limit you set before you sit down.

Playing baccarat too? Our sister site's Baccarat Predictor Tool runs the same test on baccarat's drawing rules.