Martingale Simulator
Run a doubling progression against your own base bet, table limit and bankroll, and see how often it busts.
This page takes Martingale apart on its own, spin by spin against the bet size, table limit and bankroll you set: it prices a doubling progression against the bet you would actually be allowed to place, and says nothing about the underlying game beyond the single win probability you type into it. To compare it against five other systems on the same wheel, D'Alembert, Fibonacci, Paroli, Labouchère and flat betting, the roulette strategy simulator runs all six side by side.
Set by these settings, before a bet
Most a session can win+$50
Most a session can lose−$255
Edge on every bet2.8%
The ladder a losing run climbs
From a standing start, 8 losses in a row cost $255. The next bet would be $256, over the $250 table limit.
Across 5,000 sessions
9.0%
could not place the next bet
The most the progression can be down when that happens: −$255
A link to this page opens the simulator on the same base bet, table limit, bankroll, win rate, rounds per session and profit target shown here. It carries those inputs and nothing else: nothing is stored, and nothing in it identifies you.
What 5,000 sessions pin down
Every figure below is an estimate from those 5,000 sessions, and the line under each one is the range this sweep pins it to at 95% confidence. The two rates above have them too: the bust rate at 8.2% to 9.8% and the share finishing ahead at 86.9% to 88.7%. A wide range is not a fault in the model. It is how much 5,000 sessions can settle, and the average result is always the loosest of them, because the rare deep losses that set it turn up in some sweeps and not others.
- Bust rate
- 9.0%
- 8.2% to 9.8%
- Sessions in profit
- 87.8%
- 86.9% to 88.7%
- Average result
- −$3.25
- −$5 to −$1
- Median result
- +$23
- exactly +$23
Doubling moves risk into a fat tail, it does not move the center: the average result is whatever a session stakes multiplied by the 2.8% edge carried by every individual bet, so it stays negative however the progression is arranged.
Two settings compared
More headroom does not make the progression safer, it makes the rare loss bigger.
| $250 limit, $500 bankroll | $1,000 limit, $2,000 bankroll | |
|---|---|---|
| Bust rate | 10.1% | 2.7% |
| Average result | −$6.21 | −$8.18 |
| Deepest loss (not sampled) | −$255 | −$1,023 |
A US double-zero wheel is 18 of 38, entered as 47.4%, and it moves every figure in the panel.
| 48.6% win rate | 47.4% win rate | |
|---|---|---|
| Edge on every bet | 2.8% | 5.2% |
| Average result | −$6.21 | −$11.66 |
Both tables average 10 independent 5,000-session sweeps (50,000 sessions in total) at fixed seeds, rather than tracking whichever seed the live Re-run button lands on.
Press your luck
You’re running the Martingale by hand. Each spin is a 48.6% coin flip: win and you pocket the bet, lose and it doubles. Cash out before the doubling breaks the table limit.
P/L this run
+$0
Next bet
$1
Spins
0
This run, newest spin first
No spins yet. A win adds the bet, a loss takes it and doubles the next one.
Written and reviewed by Alexander Korsager · last reviewed September 26, 2026 · corrections
The idea
What this tool shows
The Martingale doubles your stake after every loss so that one win recovers all prior losses plus one unit. It wins small amounts most of the time, which makes it feel unbeatable, until a losing streak arrives.
Run a session and watch the bet size climb: 1, 2, 4, 8, 16… A run of eight losses pushes the next bet past the table limit or your bankroll, and a single bad session erases every earlier gain. The house edge never changed.
FAQ
Questions players ask
Does the Martingale system work?
Not in the long run. It trades many small wins for rare catastrophic losses, and table limits guarantee you eventually can't place the bet needed to recover.
How many losses in a row does it take to bust?
From a $1 base with a $250 table limit, eight losses in a row ends it: the ninth bet would be $256, over the limit. On an even-money bet at 48.6% a run of eight turns up about once in every 200 progressions, which works out to roughly one busted session in ten over the 50 rounds this simulator runs by default.
Does it change the house edge?
No. Every individual bet carries the same edge; the progression only reshapes the distribution of outcomes.
Methodology
How this is calculated
The panel is in two halves and they are computed in completely different ways. The three figures directly under the fields, the most a session can win, the most it can lose and the edge on every bet, are closed forms read straight off the settings.
They are fixed before a bet is placed and they do not move when you press re-run, which is why they sit above the results rather than in them. Everything below them is measured.
The tool plays 5,000 independent sessions against the base bet, table limit, bankroll, win rate, round cap and profit target in those fields, doubling the stake after every loss and dropping back to the base after every win, then reports what those 5,000 sessions did.
A session ends when the round cap runs out, when a profit target is reached, or when the next double is more than the table allows or the bankroll covers. The sweep is seeded, so the same inputs redraw the same 5,000 sessions until you press re-run.
The figure that ties the two halves together is the one the panel does not print at all, the average amount staked per session: at the shipped defaults that is about $196 against an average result of $5.48 down, and $5.48 is 2.8% of $196, which is the edge carried by every individual bet inside it.
Common misreadings4
- The share of sessions finishing in profit, about 87% at the defaults, is not an 87% chance of coming out ahead over time. It is the same 87% every session, and the other 13% pays for all of them: the middle session wins $23 while the worst loses $255.
- The bust rate does not describe a session that felt different on the way. A busted session and a $23 winner are identical until the eighth loss lands, which is the reason the system reads as steady rather than as a run of small wins with the losses stacked at the end of it.
- More headroom does not make the progression safer, it makes the rare loss bigger. The headroom comparison above renders what raising the table limit to $1,000 with $2,000 behind it does: the bust rate falls, and in the same move the average result and the worst session both get bigger.
- A profit target does not get around the edge either. Stopping at $10 ahead cuts the average staked from $196 to $86, which lifts the average result to a $2.42 loss. It moves toward zero because the session is shorter, and it does not cross it.
Assumptions6
- Every round is an independent even-money bet won with the probability in the Win % field. The 48.6% the page opens on is close to the even-money chance on a single-zero wheel, 18 of 37, or 48.65%. A US double-zero wheel is 18 of 38, entered as 47.4%, and the wheel comparison above renders what typing that in does to every figure in the panel: a bigger edge and a bigger average loss, computed rather than quoted.
- A win pays the stake, so a progression that loses any number of times and then wins nets exactly one base bet, whatever the losing run cost on the way down.
- The check on whether a bet is affordable runs before the round rather than after it. A session refused its next bet keeps what it had rather than losing the bet it could not place, which is why the deepest loss at the defaults is the $255 of eight bets actually placed, and not the $511 it would be if the refused $256 were charged as well.
- The most a session can lose is the deepest ladder the settings allow, and it counts money banked earlier in the same session rather than the bankroll alone. A progression nets exactly one base bet however deep it went, so a session that has already completed a few of them can fund a longer ladder than a standing start can. At the shipped defaults the distinction does not bite, because the table limit caps the ladder at $255 and a $500 bankroll covers it either way. At a $20 bankroll it does: a standing start stops $15 down, and the figure printed is $20, which is what banking eleven completed progressions and then losing five bets in a row costs.
- A profit target moves the ceiling and the floor both, and the ceiling does not land on the target itself. It becomes the first whole number of base bets that reaches it, because the bank stands at a multiple of the base bet each time a progression completes and the session stops the round the target is met, so a $10 target on a $3 base prints $12 rather than $10. The floor gets shallower, because a session still running is by definition below its target, which limits how much can be standing when its last ladder starts: at a $20 bankroll a $10 target takes the most a session can lose from $20 to $15.
- The bankroll is the money brought to the session and the table limit holds for the whole of it. Neither is topped up from outside, and nothing is carried in from a previous session.
Limits4
- Five thousand sessions is a sample, not the answer. This distribution has a long left tail, so the average result is the figure that moves most between re-runs: the exact value at the defaults is a loss of $5.48, and a single sweep is only held to landing within about $3.40 of it. The bust rate and the median barely move.
- The win probability is the only thing the underlying game contributes. A bet that does not pay even money, and a rule that returns half the stake when zero lands, are different calculations, and neither is modeled here.
- The round cap is a stopping point, not a property of the game. Sessions are independent and nothing carries between them, so this says what one bounded session does 5,000 times rather than what a progression run across a year of visits does.
- Only the doubling ladder is modeled. A progression that steps by less than a double, that raises after a win instead of a loss, or that runs across bets at different odds is outside what this computes.
Evidence
The progression, the refusal rule and the sweep are in math.ts, and math.test.ts checks them against an exact enumeration rather than against the simulator's own output.
At a fixed base bet a session is a small Markov chain over two integers, the losses in the progression running now and the progressions already finished, so the true distribution of final results is computable in closed form.
That reference is where the 9.86% bust probability and the $23 median come from, and the 5,000-session sweep is required to land within three standard errors of them.
The ceiling and the floor above the simulator are not sampled at all, so nothing about them is held to a tolerance: the same enumeration says exactly how far a session can run in each direction, and the ceiling and floor printed have to match it on every setting tested, including the $20 bankroll case where the ladder is funded partly out of money banked earlier in the session.
The sweep is then held separately, on ten seeds, never to finish above that ceiling or dig below that floor.
The relation the whole page rests on is pinned separately: Wald's identity says the average result is the average amount staked multiplied by the per-bet edge, however a session happens to end, so no arrangement of stake sizes moves the center.
A passing test means the code matches that model, not that a particular table matches the numbers you typed.
Gambling can cause serious harm. The National Problem Gambling Helpline is free, confidential and open 24/7, call or text 1-800-MY-RESET (1-800-697-3738), or chat at ncpgambling.org/chat.
Read up
Go deeper
Related tools
Try next
- Roulette Strategy SimulatorRun Martingale, D'Alembert, Fibonacci, Paroli or Labouchère over many sessions, watch what happens.
- Variance SimulatorWatch a downswing play out
- Roulette PayoutsLook up any bet's true odds
- Risk of Ruin CalculatorFind the odds of going broke
- Crash Auto-Cashout SimulatorSimulate a Crash session
- Problem Gambling Screening (PGSI)The nine-question PGSI, scored as published, with support resources beside your result. Nothing you enter is saved.