Odds & House Edge

Almost no casino publishes this page. I think that is the whole problem.

Below is the exact house edge of every game on Ganbate, taken from the code that actually runs the bets — not from a marketing sheet. Two games (blackjack and video poker) have no paytable constant that sets their edge, because theirs depends on how you play; for those I computed the number exactly rather than guessing at it. Where a number in the code was wrong, I say that too — there were two, both found by writing this page, and both are now fixed. They are still described below, because a page that quietly deletes its own errata is not one you should trust with the numbers that matter.

Ganbate is free to play and the Coins are not money. You cannot withdraw them, cash them out, redeem them, transfer them, or trade them. Nobody here can lose money they cannot afford, because the Coins are not money. So why publish odds at all?

Because you should not have to take my word for it — even here. Ganbate sells Coin packs. That means a dishonest operator in my position would have an incentive to quietly tune the odds, drain balances faster, and sell more packs. That is a real conflict of interest and I am not going to pretend it does not exist. The answer to it is not a promise. The answer is that every round is provably fair and every house edge is published, so you can check both yourself.

How to read these numbers

RTP ("return to player") is the fraction of everything you wager that comes back to you, averaged over a very large number of bets. House edge is the rest: house edge = 100% − RTP.

A 1% house edge means that over many bets you should expect about 99 Coins back for every 100 Coins you wager. Not per bet — per total amount wagered. If you bet 100 Coins a hundred times, you have wagered 10,000 Coins, and the expected cost is about 100 Coins, however the individual bets went.

Three things this number is not:

• It is not a prediction of your session. The edge is a long-run average. Over a hundred bets, luck dominates completely. Over a hundred thousand, it does not.
• It is not a fee you can see. It is baked into the payout. A coin flip that pays 1.98× instead of 2.00× has a 1% edge; nothing is ever deducted, the win is just slightly too small.
• It is not one number per game. Baccarat's banker bet and baccarat's tie bet are the same game with wildly different edges. Read the table.

Every live game

House edge and RTP for a single wager, as computed from the source. Games where the edge compounds across a run are marked — for those, the headline number is not what will actually happen to your balance, so read the section below them.

Baccarat and blackjack also offer side bets, and several of them cost far more than the main game. They get their own section below.

Every casino game accepts a zero stake. The minimum bet is 0 Coins — dice.ts:19 reads "0 = free play: the full game, provably fair, nothing staked." You can play any of them, provably fair, without staking a thing. The maximum is 100,000.00 Coins. Poker is the exception: it has blinds and a buy-in, so a seat there costs at least something.

A note on Dice

Dice looks like a flat 1% game and is not quite one, and the reason is worth a paragraph because it is the kind of detail that usually gets buried.

Your roll is an integer from 0 to 10,000 — displayed as 0.00 to 100.00 — so there are 10,001 possible rolls. But your win chance is priced against 10,000. Betting under 50.00 wins on 5,000 of them; betting over 50.00 wins on 5,000 of them; and the roll landing exactly on your target loses in both directions. One outcome in 10,001 is always the house's, structurally the same way the zero pocket works on a roulette wheel.

So the true return is 9900 / 10001 = 98.9901%, and the edge is 1.0099%, not the round 1% the code comment intends. It is a hundredth of a percentage point. I am telling you anyway, because a page that rounds in its own favour is not a page you should trust with the numbers that matter.

A note on Limbo

Limbo is almost flat, and I would rather give you the "almost" than the round number.

The payout is floor(99e8 / draw) — an integer division — so the edge is exactly 1.0000% only when the target divides cleanly. It does, to four decimal places, at every target up to about 100×. Above that the remainder thrown away by the floor grows relative to the payout, and the edge drifts: worst 1.0099% near the 10,000× ceiling, best 0.9998%.

So Limbo is a 1.00% game anywhere you are realistically playing it, and a 1.01% game if you are chasing four-figure multipliers. It is not, strictly, "identical at every target" — an earlier version of this page said it was, and that was wrong.

A note on Crash

Crash is built from the same 0.99 as everything else, but it does not end up at a flat 1% edge, and the reason is worth stating exactly.

The round crashes on the same millisecond the curve reaches its crash point, and the server classifies that millisecond as crashed — so a cash-out at x pays only when the crash point is strictly above x, not merely equal to it. That makes the true expectation of a target x:

EV(x) = x × 0.99 / (x + 0.01)

which approaches 99% from below and only reaches it in the limit: 1.97% edge at 1.01×, 1.89% at 1.10×, 1.49% at 2.00×, 1.01% at 100×. Low targets are the expensive ones — and the low targets are exactly where a nervous player cashes out. Bailing at 1.01× costs you nearly twice what the headline 1% suggests. 99.00% is a ceiling, not an average.

The other number people get wrong is how often it busts instantly.

The crash point is max(1.00, 0.99 / f) for a uniform f (crash.ts:22-26). That gives P(point ≥ x) = 0.99 / x, so the chance of reaching even 1.01× is 0.99 / 1.01 = 98.0198%. Everything below that floors to 1.00× and the round dies on the spot:

1 − 0.99 / 1.01 = 1.9802% of rounds bust at 1.00×.

Not 1%. Roughly one round in fifty — about twice what the headline edge might lead you to guess, and it is a natural mistake to make, because the two numbers look like they should be the same. They are not. The edge is what the game costs you on average; the instant bust is the single most likely way a round ends. Keep them apart.

A note on Plinko: where our paytables differ from the standard ones

Plinko is 27 separate paytables — 3 risk levels × 9 row counts — and the edge runs from exactly 1.0000% (medium/9 and high/8, the cheapest bets in the game) to 1.09375% (medium/8 and medium/10, the dearest). Every one of the 27 is enumerated on every build.

The tables themselves are the industry-standard ones: the same prices the market leader prints, for every row count from 8 to 16. Fourteen of the 27 are byte-for-byte identical to the standard tables. Thirteen are not, and you are entitled to know which and why.

The standard tables target 99% RTP and round every price to the nearest 0.01×. That rounding lands thirteen of them a hair above 99% — high/11, as published, is a 99.1602% table. An operator with no Zero-Edge allowance can ship that without a problem. We cannot, because Edge Rebate refunds the edge by scaling a payout by 100 ÷ 99, which only stays at or below break-even when the edge really is at least 1%. Under Edge Rebate a 99.16% table pays over 100% — a bet the house loses money on, farmable to the daily cap. That is not a hypothetical; it is precisely what happened here with high/8 and high/12.

So thirteen tables carry the smallest correction that restores the 1% floor: one bin moves, and never the jackpot. It is always bin 1 — the second-from-edge bin — so the headline multiplier, the centre floor and the whole interior stay exactly as published, and the bin that moves goes to the largest 0.01× price that still clears the ceiling. That is why high/8 pays 3.99× rather than a round 4×, and high/12 pays 23.8× rather than 24×.

The full bin-by-bin divergence — all thirteen, with the standard price beside ours — is published on the Plinko rules page. It costs you at most 0.16% of RTP on a single table, and it is what makes an allowance that hands the edge back to you affordable. We would rather print the asterisk than quietly run a game the house cannot afford.

Games where the edge compounds

This is the most important section here, and the one a casino with something to hide would leave out.

Tower, Chicken and HiLo charge the house edge on every step, not once per round. Each safe row, each lane crossed, each correct guess multiplies your payout by roughly 0.99 ÷ P(surviving that step). That 0.99 is the 1% edge — and it is applied again every time you continue.

So it compounds:

A chicken that crosses all twenty lanes has played against an 18.3% house edge on that run, not 1%. A full nine-row tower climb faces 8.7%. These are not estimates; they come straight from the multiplier code, and the repository's own tests say it out loud (tower.test.ts:69 — "The 1% per-step edge compounds: fair EV at k rows is 0.99^k").

Mines does not work this way, and the contrast is worth understanding. In Mines the 0.99 is applied once, to the whole run's survival probability: multiplier = 0.99 × C(25,k) / C(25−m,k). The consequence is elegant. Stopping after one tile and stopping after fifteen face the same 1% edge — and every additional tile you reveal is a mathematically fair, zero-edge decision. You pay the 1% when you place the bet, and never again.

One asterisk, because the table above carries it: multipliers are truncated to two decimals, and on a small multiplier that truncation is a bigger slice of a smaller number. The worst cases are the cheap early cash-outs — 1 mine and 4 reveals pays 1.17× where exact arithmetic wants 1.1786×, a 1.72% edge rather than 1.00%. It is the only place Mines departs from flat, and it departs against you, so it belongs in daylight rather than in a footnote.

I am not going to claim the compounding games are built that way for your benefit. They follow the genre standard, and the genre standard is expensive. At least now you know which is which — and that in Tower, Chicken and HiLo, the cheapest move available to you is always to stop.

The two numbers that were wrong

Writing this page is what found them. Both are fixed in the code now, and I am leaving the account here rather than deleting it — an odds page you can only trust when it has nothing to admit is not much of an odds page.

Wheel, medium risk, had a 2% house edge

Every other game on this site targets 1%. Wheel's medium layout used to be [165, 0, 200, 0, 300, 0, 150, 0, 165, 0]. Those multipliers sum to 980 rather than 990, which made the RTP exactly 98.00% — a 2% edge, double the rest of the casino, at every segment count. The file's own header described the design as a "layout whose mean is ≤ 0.99 (1% house edge)", so the code and the comment disagreed, and the comment was the one telling the truth about the intent.

It slipped through because the property test only asserted that the expected value landed somewhere between 0.95 and 0.99 — a band roomy enough for a 2% edge to sit in unnoticed.

Fixed. The 150 is now a 160, the layout sums to 990, and medium risk is a 1% game at every segment count like everything else. The property test now asserts EV === 0.99 exactly, for all nine (risk, segments) layouts, so the band that hid this cannot hide it again. See wheel.ts and wheel.test.ts.

Baccarat's pair side bets were priced for the wrong deck

All four pair bets were inherited from the eight-deck table game and were badly mispriced on Ganbate's single deck, where every one of those events is far more likely. Three were brutal against the player and one was a giveaway:

A 29% and a 32% house edge on the pairs — and a player-positive perfect-pair bet, at a −0.92% edge, which anyone betting it exclusively could have printed Coins on.

Fixed. All four are repriced to the same ~1% edge as the rest of the site, and baccarat.test.ts now pins each one. The probabilities are exact, not sampled: a pair bet reads only the four opening cards, so the test enumerates every opening a single deck can produce — all 1,624,350 of them — and checks the price of each bet against it.

I am telling you about the one that was in your favour as loudly as the ones that were in mine, because a page that only discloses errors it profited from is not an honest page.

Side bets: read this before you touch them

Side bets are where a real casino makes its money. Ganbate makes nothing from them: the Coins you wager are not money, and the edge feeds nothing but the Coin sink. (What Ganbate does sell is Coin packs — which is exactly the conflict of interest I opened this page with, and exactly why the numbers below are published rather than promised.)

Side bets are usually where the edge is steepest by a wide margin. On Ganbate's baccarat they mostly are not — not any more — though two of them still are, and those two are worth seeing before you place one. On Ganbate's blackjack they very much are. Every figure below is exact, from exhaustive enumeration of the code's own paytables.

Baccarat (single deck):

The four pair prices (16.83× / 25.50× / 8.67×) look strange next to the 11:1 and 25:1 you may know from an eight-deck table. They are strange on purpose: on a single deck those events are far more likely, so the eight-deck prices were wrong in both directions. They are now priced to this deck, at the same ~1% edge as the rest of the site. (What they used to be, and what that cost, is in the section above.)

If you take one practical thing from this page: do not bet the tie, and think twice about the banker bonus. Everything else on the baccarat felt costs you about 1%.

Blackjack side bets (Perfect Pairs, 21+3) are dealt from the six-deck shoe with the standard tiers (blackjack.ts). Unlike the blackjack hand itself, they do not depend on how you play: they settle the moment the cards hit the felt, off your first two cards and — for 21+3 — the dealer's up-card. That makes them finite and countable, so there is no excuse for not knowing what they cost. Enumerating every deal the shoe can produce (all 48,516 two-card openings, all 5,013,320 three-card ones):

Perfect Pairs — settles on your first two cards. RTP 93.8907%, house edge 6.1093%.

21+3 — settles on your first two cards plus the dealer's up-card. RTP 95.3790%, house edge 4.6210%.

Insurance is the third fixed-price bet blackjack offers, and the most expensive of the three. It is not placed with the main wager — the dealer offers it, for half your stake, whenever their up-card is an ace — but it prices like any other bet, so it belongs here. It pays 2:1 that the dealer's hole card is worth ten. On a six-deck shoe, once the dealer's ace is removed, the hole card is a ten with probability exactly 96/311 = 30.87%, so a bet paying 2:1 returns 3 × 96/311 = 288/311 = 92.6045% — a 7.3955% house edge. That makes insurance the worst bet at the blackjack table: dearer than Perfect Pairs, dearer than 21+3, dearer than any decision inside the hand itself. It wins only when the hole card is a ten more than one time in three, and the best a player can ever see — holding no ten in their own two cards — is 31.07%, still short of the 33.33% it takes to break even. So optimal play never insures, and even money (the same bet offered against your own natural) is insurance wearing a friendlier name. The felt offers it with no warning. This page is the warning.

Those are the standard six-deck tiers, and the standard six-deck tiers are expensive. Against the ~1% the rest of this site charges, Perfect Pairs costs about six times as much and 21+3 about four and a half. Among the one-shot bets on this site that carry a fixed price, only the baccarat tie (15.75%), the baccarat banker bonus (9.74%) and blackjack's own insurance (7.40%) cost you more than these two.

I have not repriced these to 1% the way I repriced the baccarat pairs. That is a choice, and you are entitled to see it as one: these are the tiers the game is known by, and they are correct for a six-deck shoe rather than wrong for it. But correct is not the same as cheap. Perfect Pairs takes 6.11% and 21+3 takes 4.62%, every single time you place them. They are printed here so that if you bet one, you bet it knowing.

The two games whose edge depends on how you play

For thirteen of the sixteen games the edge is a constant I can point at in the source. For two — blackjack and video poker — there is no such constant, because the edge is not a multiplier: it emerges from the rules and from how well you play. (Poker is the sixteenth, and it has no house edge at all; it has rake. See below.)

That does not make the number unknowable. It makes it something that has to be computed rather than read off a line of code — an exact evaluation of every decision against every deal the shoe can produce. I have now run both computations, and both numbers are below. They assume optimal play; play worse and you pay more, with no ceiling on how much more.

Blackjack

The edge here is made of the rules, so here they are exactly, from blackjack.ts and service.ts:

• Six-deck shoe, freshly shuffled every single hand.
• Dealer stands on all 17s, soft 17 included (dealerShouldDraw).
• Blackjack pays 3:2 — 2.50× your stake returned, not the 6:5 that has become common on modern casino floors.
• Double on any first two cards, including after a split: one card, then stand.
• Split any pair, and resplit to three hands in total. Split aces take exactly one card each and cannot be resplit, and a two-card 21 on a split hand is a plain 21 that pays even money — not a 3:2 blackjack (blackjackSplitHandResult).
• Late surrender — give up a hand before you have drawn to it and take half the stake back.
• Insurance (and its twin, even money) is offered when the dealer shows an ace. It is a losing bet and correct play always declines it — priced in full below.
• The dealer's hole card is checked before you act. With a ten showing the check is silent and immediate; with an ace showing the hand pauses for your insurance decision and peeks after. Either way you never double or split into a dealer blackjack.

This is close to the most generous rule set blackjack is dealt under anywhere: 3:2, stand on soft 17, split and resplit, double after split, late surrender. The two familiar house-favourable trims are both absent — the dealer does not hit soft 17, and the natural is not cut to 6:5 — and every option a stingier table takes away, split and surrender included, is on this felt.

The number is a 0.3378% house edge at optimal play — an RTP of 99.6622%, measured per Coin of your opening stake. Per Coin actually wagered — splitting and doubling put more money on the felt — it works out at 0.2986% (99.7014%); that is the figure the house's own solvency math binds on, and it is still comfortably under 1%.

That is not a textbook figure lifted from a table with different rules, and it is not a simulation. It is an exact evaluation — dynamic programming with full card-removal effects — of the rules above, calling the real blackjackResult, the real dealer-drawing rule and the real payout constants, so it cannot drift from the game without the game changing (main/tools/scripts/analysis/rtp-blackjack-rules.ts, ruleset "B"). The one place it is not closed form is the split — sibling hands draw from a shoe they have already depleted — so that script also replays twenty million split rounds under the fully exact sequential game and measures the residual: +0.0007% ± 0.0024% of RTP. The number is right to a thousandth of a percentage point. As a further check the script re-derives the old no-split game to −0.009714136, the number this page used to print, or it refuses to run.

Optimal means optimal. It is the return of a player making the best available decision on every hand. Any deviation costs more, and nothing bounds how much more.

Blackjack is, by a wide margin, the cheapest bet on the entire site — about a third of the flat ~1% edge every other game charges, and well below even the 0.97% the old no-split version ran at. The reason is worth stating plainly, because it is the whole design decision: the single option worth most to the player is the split, and it is worth more than every house-favourable trim combined. Adding split to this shoe hands the player about 0.37% of RTP. Hitting soft 17, going to eight decks, banning double-after-split, banning resplit, dropping surrender — all of those together claw back less than that one option gives. Computed exactly from the full rule matrix (rtp-blackjack-rules.ts), no 3:2 blackjack that still lets you double reaches even a 1% edge — the house-friendliest such game tops out at a 0.92% edge. You can force a split game past 1% only by gutting it: dropping double-down entirely gets there (a ~2.1% edge), but at a 1.4% cost to the player, more than three times what cutting the natural to 7:5 (a 0.4% cost) would take. Every route to a 1% edge is a worse deal for the player than blackjack's own price. So blackjack keeps its 3:2 price and stays the most generous bet in the house, deliberately. It is also, played perfectly, still a losing game.

One honest consequence of pricing it there. The casino has a single global edge control — the same payout-rescale mechanism Edge Rebate is built on, an admin-set target that every other game's payouts are scaled to. Its safety argument assumes a game is designed at or below 99% RTP: scale a 99%-designed payout up to the control's 100% ceiling and you land exactly at break-even, never past it. Blackjack is designed above that line, so the argument does not hold for it — rescaled to the same ceiling it would return 100.71% per Coin wagered, a game the house pays you to play, and it goes underwater from a target of 99.30%, thirty basis points above the default. So blackjack is exempt from the rescale entirely (RESCALE_EXEMPT_GAMES in houseEdge.schemas.ts): every blackjack round is quoted at the baseline rate and settles at its designed price, wherever the control is set, in either direction. That is a deliberate carve-out, not an oversight, and a test holds the pairing on every build — a game designed above 0.99 without the exemption fails it (zeroedge.invariant.test.ts). (Edge Rebate, the separate daily 0%-edge allowance, has never applied to blackjack either — it covers one-shot instant bets only.)

Insurance is the exception to all of this — the one blackjack wager that is genuinely expensive, at a 7.40% edge. It is priced in full in the side-bets section above, along with why optimal play never touches it.

One consequence worth stating plainly: card counting is worthless here. The shoe is rebuilt from a fresh seed at the start of every hand, so nothing carries from one hand to the next. There is no penetration to exploit and no count to keep.

Video Poker

The paytable is in the code (videopoker.ts:78-89), so I can give you that exactly. These are totals returned per 1 Coin staked — so "jacks or better" returns your stake and nothing more. It is a push, not a win.

The 9 and the 6 on full house and flush make this the 9/6 Jacks or Better schedule — historically the good one. One caveat, because anyone who knows video poker will otherwise assume too much of that name: the famously high return on 9/6 Jacks or Better assumes a max-coin royal bonus, where betting five coins lifts the royal from 250× to 800×. There is no such bonus here. The royal pays a flat 250× at any stake — and a smaller royal cannot produce a bigger return, however you play. So this game's best possible return sits below the number that name brings to mind, and the gap is quantified below. The 9 and the 6 are the good news. The royal is the fine print.

What the code does not contain is an RTP, because video poker's return depends on which cards you choose to hold — a strategy question, not a constant. So it had to be computed: every deal evaluated against each of the 32 ways to hold it, scored by the real paytable.

At optimal play this game returns 98.3735% — a 1.6265% house edge.

So it costs noticeably more than blackjack (0.34%) and more than the flat ~1% most games here charge. The gap between 98.37% and the 99.54% the name 9/6 Jacks or Better usually implies is exactly the missing max-coin royal bonus described above — the flat 250× royal is what costs you the difference, and it is the reason the familiar name promises more than this game pays.

And the spread between playing well and playing carelessly is enormous — far wider than in blackjack. A player who simply never discards, or always discards, gets back 33.6021%. Video poker punishes inattention harder than any other game on this site. If you are clicking through it without thinking about which cards to hold, you are not playing a 1.6% game; you are playing a 66% one.

The double-up gamble is exactly fair. After a winning hand you may put the round's payout back at risk: pick one of four face-down cards against a dealer card. Higher doubles it, lower loses it, a tie rides. Because you pick blind and the dealer's card is never shown to you first, your card and the dealer's are statistically interchangeable — so your chance of winning is exactly equal to your chance of losing, and ties return the pot. The expected value is 1.00: a 0% house edge. Pure variance, nothing skimmed. (I checked the server to be sure: the dealer's card is not sent to your browser until after your pick is locked in.)

Poker has no house edge — it has rake

At a poker table you are not playing the house. You are playing other people, and the Coins move between players. The house takes no position on the hand.

What it takes instead is rake: 10% of a cash-game pot, capped at 50.00 Coins (rake.ts:11-18), and only once a flop has been dealt — no flop, no drop (poker.service.ts:398-406). ("Cash game" is the poker sense of the term — a ring game rather than a tournament. There is no cash in it.) Tournaments rake at payout instead. That is not a house edge; it is a fee on Coins that have already moved between players. Your expected result at a table depends on the other players, not on a constant I control.

One thing I have to be straight about: house bots can occupy poker seats. They are seated by an admin rather than automatically, and they play a heuristic strategy. If you sit down, some of the seats may not be people.

Edge Rebate: the 1% given back

Edge Rebate is a daily allowance that refunds the standard 1% on qualifying bets. While it is active your payout is multiplied by 100 ÷ 99 (zeroedge.schemas.ts:23-25), which cancels the 0.99 on a game priced at 0.99 and lifts the return to approximately 100%.

• Applies to the eight instant games: Dice, Limbo, Keno, Plinko, Wheel, RPS, War, Baccarat.
• Per-bet limit: 1,000.00 Coins. Bigger bets do not qualify.
• Daily allowance: the first 50,000.00 Coins you wager each day, resetting at UTC midnight.
• Does not apply to the multi-step rounds (Mines, Tower, Chicken, HiLo, Video Poker, Blackjack), to Crash, or to Poker.

Two things that "approximately" is hiding, and both of them matter:

Edge Rebate refunds 1% — not whatever the bet actually costs. The multiplier is a flat 100 ÷ 99 on the payout. It is not a promise of a 100% return. On a game priced at 0.99 it cancels the edge and you land within a whisker of even. On baccarat's expensive bets it does nothing of the kind: the tie qualifies for Edge Rebate like everything else, and 84.2539% × 100 ÷ 99 is 85.10%. A refunded tie is still the worst bet in the casino. Edge Rebate does not fix a bad bet; it discounts a fair one.

And it does not round only one way — but it no longer rounds the wrong way. Payouts are floored to whole hundredths of a Coin, which shaves the remainder toward the house: dice, for instance, comes out at 99.99% rather than 100%. The opposite used to happen too, and it was a bug rather than a perk. Seven configurations across three games — plinko on high risk at 8 and 12 rows, keno at 2, 3, 8 and 10 picks, and baccarat's perfect pair — were priced a hair above 0.99, and a flat 100 ÷ 99 over-refunds anything priced above 0.99. Inside the allowance those seven returned more than 100%: a bet the house lost money on, farmable to the daily cap by anyone who read this page carefully. All seven are re-priced.

The refund is safe exactly when a game's designed RTP is at or below 0.99, because 0.99 × 100 ÷ 99 = 1.00 precisely — so that ceiling is now enumerated, not assumed. A test walks every configuration of every table-driven game on every build and fails if a single one designs above it (zeroedge.invariant.test.ts), which for Plinko alone is all 27 (risk, rows) tables. (Blackjack is the one game designed above the ceiling — 99.6622%, deliberately — which is why it is not on the Edge Rebate list and is exempt from the payout rescale altogether; the same suite proves that exemption at every setting of the global edge control.) The tightest bets on the site — Plinko at medium/9 and high/8; keno at 1 pick; baccarat's player and banker pair — sit at exactly 0.99, so Edge Rebate takes them to exactly break-even. That is the ceiling, not a step past it. Both effects now lean the same way, toward the house, and both are hundredths of a percent: neither is worth building a plan around. I am spelling this out because "approximately 100%" is doing real work in that sentence, and you should know which way each approximation leans.

What strategy can and cannot do

I want to be precise here, because "no strategy works" is usually said sloppily, and a careful reader will rightly push back on it.

Strategy chooses which edge you face. It cannot make the edge positive.

Some of your decisions genuinely move the number:

• In Baccarat, betting banker (1.01%) instead of the tie (15.75%) is a real and enormous improvement.
• In Keno, Plinko and Mines, how you configure the bet selects the edge you play against.
• In Crash, a low cash-out target is genuinely more expensive than a high one — the edge runs from about 1.98% at the very bottom (1.97% at 1.01×, 1.89% at 1.10×) down toward 1.00% at 100×.
• In Tower, Chicken and HiLo, stopping earlier really does lower your compounded edge. The optimal strategy is not to take the next step.
• In Blackjack and Video Poker, playing well is an actual skill and it is worth real percentage points — in video poker, the distance between optimal play (98.37%) and thoughtless play (33.60%) is the widest gap on the entire site.

Every one of those moves you from a worse negative number to a better negative number. None of them reaches zero, let alone crosses it.

In Dice, RPS, War and Wheel the edge is genuinely fixed: no choice you make moves it. Not the stake, not the target, not the risk level, not the timing. Betting under 2.00 and betting under 98.00 cost you exactly the same fraction, and Wheel's low, medium and high layouts all return the same 99%. Limbo is fixed for any target you are likely to pick (1.0000% up to ~100×) and creeps to 1.0099% only if you chase the 10,000× ceiling.

Keno, Plinko, Mines and Crash are different, and I have to be careful not to overstate the point. In those four, the edge depends on how you set the bet up — pick count in Keno (1.00%–1.04%), risk and rows in Plinko (1.00%–1.09%), mine count and when you stop in Mines (1.00%, up to 1.72% on small multipliers), cash-out target in Crash (1.98% down to 1.00%). Those are real choices with real consequences, and the table above is how you make them.

What no decision changes in any of these nine games is the outcome. Once the bet is placed, the result was already fixed by a hash before you touched anything. You choose which edge to face; you never choose the roll.

Betting systems cannot help anywhere. Martingale, doubling after a loss, waiting for a number that is "due", pressing a hot streak — each is just a sequence of individually negative-expectation bets. The expected value of a sum is the sum of the expected values. Choosing when to bet and how much reshapes your outcome distribution — it makes big wins and big losses more or less likely — but it cannot move the average. There is no ordering of losing bets that adds up to a winning one. That is arithmetic, not an opinion.

Past results do not affect future ones. Every bet you place draws from an independent HMAC-SHA256 stream keyed by your server seed, your client seed, and a nonce that increments once per bet. Bet 500 knows nothing about bet 499. A dice roll that has come up low ten times running is exactly as likely to come up low on the eleventh. The wheel is not "due". The crash is not "overdue for a big one". The multiplier does not remember.

(The one real dependency lives inside a single round: when Mines hides its bombs or HiLo shuffles its deck, cards come out without replacement, so what has been revealed does change what remains. That is normal, it is how a deck works, and the multipliers already price it in exactly.)

Why have a house edge at all?

A fair question. Nothing here is worth money. I could set every game to exactly 100% and let the Coins float.

Two reasons I did not.

The games would stop behaving like the games. A crash curve with no edge never busts at 1.00×. Dice at true odds pays 2.00× on a coin flip. Every intuition you have about how these games feel — the near misses, the way a long climb gets punished, the fact that the tempting bet is usually the expensive one — comes from the edge. Take it out and you are not playing a simulation of the thing any more; you are playing something else that looks like it. If the point of a free casino is to let you experience these games without the harm, the experience has to be real. Which means the edge has to be real — and visible.

The Coin economy needs a drain. Coins pour in constantly: 50,000 to start, 1,000 free every day, 200 when a rescue tops up an empty wallet, 200 for a rewarded ad. If nothing ever removed them, balances could only grow, every number on the site would inflate toward meaninglessness, and a balance would stop being worth looking at. The house edge is the sink that holds the whole thing in equilibrium. It is not extracting value from you — there is no value to extract. It is keeping the numbers honest.

And the reason I care about most: an edge you can see is an argument. The single most useful thing this site can teach is that the house always wins in the end, and that it wins by a margin small enough to feel like nothing and large enough to be certain. One percent sounds like a rounding error. Ten thousand bets later, it is not.

What a win here does not mean

This is the part I would most like you to actually read.

A winning streak on Ganbate tells you nothing about your odds anywhere else. Not because our games differ from real ones — because they are the same games, and the same games are negative-expectation for the player everywhere they are offered. You did not beat a system. You had a good session inside a system that is designed to hand out good sessions on the way to a certain outcome.

Practising here will not improve your real-money odds. In Dice, Limbo, Crash, Keno, Plinko, Mines, Wheel, RPS, War, Tower, Chicken and HiLo there is no skill to build. The outcome is decided by a hash before you act, and no amount of repetition moves it. There are choices in some of them — which pick count, which risk level, when to stop — but those are things you can simply be told once, from the table above. They are not a craft. They do not improve with practice, and the best of them still loses.

Blackjack and Video Poker are the honest exception: the decisions in them are real, they are learnable, and what you learn here would work at a real table. What it would not do is make that table profitable. It takes you from a bad negative number to a less bad negative number, and any confidence you build here that it might do more than that is confidence built on a false floor.

Real gambling has one thing Ganbate does not: your money, gone, and not coming back. If a session here felt good, the honest conclusion is that the feeling is cheap to manufacture — not that you are the exception.

If any of this is starting to feel less like a game and more like something else, the responsible-play tools are real: a rolling 24-hour loss limit that tightens the moment you set it and only loosens after a 72-hour wait, and self-exclusion for 24 hours, 7 days, 30 days or permanently — irreversible for its duration once set. The National Problem Gambling Helpline (US) is 1-800-GAMBLER.

Check my work

You should not have to believe a single number on this page.

Every casino outcome comes from HMAC-SHA256 over a server seed we commit to in advance (we publish its hash before you bet), a client seed you control, and a nonce that steps once per bet. Rotate your seed pair on /fairness and the old server seed is revealed to you; paste it into /verify and you can recompute the outcome yourself, in your own browser, byte for byte. All sixteen games are covered by the verifier, poker and blackjack included.

Two of them work a little differently, and you should know how:

• Crash is one shared round for everybody, so it cannot use your personal client seed — a round belongs to every player at once. Each round gets its own server seed, its hash is published before betting opens, and the seed is released the moment the round busts. Commit-and-reveal, verifiable after the fact, but you cannot inject entropy into it.
• Poker uses a separate, versioned fairness spec from the casino games (pf-hmac-sha256-fy-v1), with the table id as the client seed and the hand number as the nonce.

The odds math itself lives in main/shared/src/modules/core/casino/ — one small file per game, all of it plain arithmetic, all of it readable in an afternoon. Every number on this page was produced from those files exactly, not sampled from them: exhaustive enumeration where the outcome space is finite (every keno pick, every plinko path, all 1,624,350 baccarat openings, all 2,598,960 video-poker deals), and exact dynamic programming with card-removal effects for blackjack, where the edge depends on your decisions. The scripts that do it are in main/tools/scripts/analysis/ and they call the same functions the server calls, so they cannot quietly disagree with the game. Where a simulation appears at all, it is only there to cross-check the exact answer.

If you find a number here that does not match the code, tell me: [email protected]. I am one person, not a support team — I read everything and reply as soon as I can, but it may take a few days. If a number on this page is wrong, I will correct it. Both of the errata above got here that way.

Related: How it works · Verify a bet · How Coins work · Your bets · Responsible Play

Last verified against the codebase: August 2, 2026. Ganbate is a free-to-play social casino for adults 18+, operated by Abraham Ahn, sole proprietor, Abraham Ahn, 5013 S Louise Ave, Unit #972, Sioux Falls, SD 57108, USA. Coins have no cash value. They cannot be withdrawn, cashed out, redeemed, transferred between players, gifted, sold, or traded. There are no prizes of monetary value. No real-money gambling takes place on this site. Contact: [email protected].