The D’Alembert system tells you to raise your stake after every loss and lower it after every win, on the theory that wins and losses must even out. It is named after an eighteenth-century mathematician who believed exactly that, and who was wrong about it.
D’Alembert is the gentlest of the four staking systems players bring to crash games, and the last one this series takes apart. It moves in single units rather than the doubling of Martingale or the leaps of Fibonacci, so the stakes climb slowly and the sessions feel calmer. That mildness is the only thing that makes it different.
The verdict is identical to the other three. No way of sizing or ordering your bets changes what a negative-edge game returns, and a system that tells you to bet more after losing simply puts the most money on the table at the worst possible time. What sets D’Alembert apart is where its logic comes from: a genuine probability mistake made by the man whose name it carries.
The verdict
D’Alembert raises your stake by one unit after a loss and lowers it by one after a win. It is the mildest staking system, with the lowest peak stakes and the slowest path to ruin, but it shares the single fatal verdict of all four: it cannot beat the house edge or even dent it. Over any run of rounds you expect to lose 3% of everything you stake, and raising bets after losses only commits more money at the worst moment.
🧠 The man who got the maths wrong
The D’Alembert system is named after Jean le Rond d’Alembert, and the irony is that he is the worst possible namesake for a betting strategy. d’Alembert was a first-rank mind: co-editor of the Encyclopédie alongside Diderot, author of the wave equation and d’Alembert’s principle in physics, and Permanent Secretary of the Académie française from 1772. On one specific question he was simply, demonstrably wrong, and it happens to be the exact question this system depends on.
In his 1754 Encyclopédie article Croix ou Pile, which translates as Heads or Tails, he argued that the probability of two heads in a row was not one in four. He reasoned that once a head appeared there was no need to toss again, treated the outcomes as heads, then tails-followed-by-heads, then tails-followed-by-tails as if they were equally likely, and arrived at two-thirds instead of the correct three-quarters. He went further in his Opuscules mathématiques of 1761: asked for the chance of tails after three heads in a row, he insisted it must “obviously” be greater than one half. It is not: each toss is independent and the chance stays one half.
📖 Definition
The gambler’s fallacy is the belief that past independent results change future probabilities, that a run of one outcome makes the opposite “due.” For independent events there is no such force. A coin, a roulette wheel and a crash round carry no memory of what came before.
That belief is the entire foundation of the system that bears his name. Its selling point is that wins and losses must balance out, so staking more after losses and less after wins keeps you near break-even and recovers your losses with smaller, higher-staked wins. It assumes the past pulls the future back toward correction. It does not.
📝 For the record: d’Alembert did not design the staking plan himself. Historians place the betting system in the late eighteenth or early nineteenth century, loosely built on his probability speculations rather than written by him. The fair statement is that the system embodies his error, not that he sat down and invented it.
What survives all the history is one blunt irony: it is the one betting system named after the man whose mistake it repeats.
“It is the one betting system named after the man whose mistake it repeats.”
⚡ How the D’Alembert system works
The rule is the simplest of any staking system: move your stake one unit up after a loss and one unit down after a win. You never drop below your starting unit, which acts as a floor, and the natural cash-out target is 2.00x, where a win pays even money. It is a negative progression, which means your bets rise precisely when you are losing.
- After a loss, stake one unit more. A loss at two units means the next bet is three.
- After a win, stake one unit less. A win at three units means the next bet is two.
- Never go below the base unit. The starting stake is the floor; you sit there until the next loss.
With a one-pound base unit, a typical run reads: bet 1, lose, go to 2, lose, go to 3, win, drop to 2, win, drop to 1, and hold at 1 until the next loss. The escalation is linear and gentle. The promise attached to it is that a losing streak gets repaid by a later winning streak at higher stakes, so you climb back to break-even with fewer wins than losses. That promise is the gambler’s fallacy in operational form, and the next section is why it cannot pay off.
🔢 Why no staking system can win
No staking system can beat a negative-edge game, and the reason is arithmetic rather than opinion. At a 2.00x target on a 97% game, each round wins 48.5% of the time, which is 0.97 divided by 2, and returns minus the house edge on each stake on average. That per-round figure does not move when you make the stake larger or smaller, so by linearity of expectation your total expected loss is just the edge applied to everything you stake, whatever order the wins and losses arrive in.
We set out the full derivation, including variance and why every system fails, in our crash gambling maths guide. The one-line conclusion is all this article needs: expected value equals minus the house edge times your turnover, at every cash-out target, under every system.
This is settled ground, not a CrashEdge opinion. Richard Epstein established the constant negative expectation of commercial games decades ago; Stewart Ethier formalised it as the conservation of fairness and named D’Alembert explicitly among the six systems it defeats; and Joseph Doob’s optional stopping theorem shows that no rule for varying stakes, and no rule for choosing when to walk away, can turn a losing game into a winning one. The Wizard of Odds states the practical upshot without varnish: “Like all betting systems, not only can’t it overcome the house edge, it can’t even dent it.”
📊 A 20-round session, in numbers
The clearest proof is to run the system and read the result. The table below walks a plausible, slightly unlucky session of eight wins and twelve losses across twenty rounds at a one-pound base unit and an even-money cash-out target, tracking the running stake, each round’s profit or loss, the cumulative amount staked and the running bankroll.
Total turnover is 59 pounds, so the mathematically expected loss is 3% of that, which is 1.77 pounds, regardless of the order of results. The realised result here was minus 7 pounds, driven by the closing run of losses across rounds 18 to 20 that pushed the stake up to 5 pounds exactly as the losses landed. That is D’Alembert’s failure mode in a single snapshot: the system stakes the most at the precise moment you are losing.
The bottom line: Flat betting the same twenty rounds at one pound stakes just 20 pounds, for an expected loss of 60 pence. D’Alembert nearly tripled the turnover and tripled the expected loss to arrive at the identical place, a slow bleed at the house edge. More money changed hands for no change in the odds.
That failure mode is not only a maths problem. A system that instructs you to bet more the moment you fall behind is engineering the exact behaviour that turns a bad session into a damaging one.
Raising your stake the moment you fall behind is the textbook shape of chasing losses, one of the most studied risk factors in gambling harm. We cover the research, the warning signs and where to find help in our guide to crash gambling and player harm.
◆
🏆 The mildest of the four systems
D’Alembert is unmistakably the gentlest negative progression, and that is its only real distinction. Linear single-unit steps climb far more slowly than Fibonacci’s additive sequence or Martingale’s doubling, which means lower peak stakes, a slower path to ruin and calmer sessions. The trade is that recovery is correspondingly weak: because every win cuts your stake, clawing back a losing streak needs a comparably long winning streak, where Martingale needs only one win.
Set against the other three, the picture is consistent. Martingale escalates fastest and most dangerously, Fibonacci sits in the middle, and Anti-Martingale simply reverses the direction by chasing winning streaks instead. Every one of them lands on the same expected loss. There is no winner in that table, only different shapes of the same bleed.
The difference in escalation is stark after a bad run. Starting from one pound, a ten-loss streak leaves each system in a very different place for the next bet:
- D’Alembert: next bet 11 pounds. You would have staked 55 pounds across the ten losses. The stake creeps, it does not spike.
- Fibonacci: next bet 89 pounds. Cumulative stake across the ten losses is 143 pounds.
- Martingale: next bet 1,024 pounds. Cumulative stake is 1,023 pounds, and a single table limit ends the run.
That gentleness is exactly why D’Alembert appeals, and exactly why it is the slowest to recover. You accept a lower chance of hitting a session target in exchange for a quieter ride. The destination is unchanged.
⚙️ Why you cannot automate it
D’Alembert cannot be run on a crash game’s auto-bet panel, because its defining step is additive and the panels are not built for additive steps. On the standard implementations the market splits into two camps, and neither can encode a fixed plus-one or minus-one unit move.
- Branded games like Aviator, Spaceman and JetX do not adjust the stake at all. Their auto-play simply repeats the same fixed stake for a set number of rounds, bounded by stop-loss and stop-win limits. There is no on-win or on-loss staking control of any kind.
- Originals-style panels like Stake and BGaming adjust only by percentage. They offer an on-win or on-loss “reset or increase by a percentage” toggle, which is multiplicative. A percentage suits Martingale, where doubling is a flat plus-100%, but it cannot hold a fixed one-unit step, because going from 3 to 4 units is a 33% rise while 4 to 5 is only 25%.
The consequence is the same one that defeats Fibonacci: D’Alembert needs a manual stake adjustment every single round, or a clumsy percentage approximation that drifts away from the true rule. There is no native way to automate it.
🛡️ Why it fails in crash games specifically
Crash rounds are independent, which removes the one thing D’Alembert needs to be true. The crash point is fixed before betting closes, determined on its own and not in relation to any previous round. In provably fair titles such as Aviator and Stake Crash it comes from a server seed, your client seed and a nonce run through a cryptographic hash, and you can verify it after the fact. In certified-RNG titles, an audited random number generator does the same job.
Because nothing connects one round to the next, there is no equilibrium force pulling results back toward balance, and no result is ever “due.” That is the gambler’s fallacy applied to a multiplier that has no memory, by a system named after a coin that had none either. The premise that a loss makes a recovering win more likely is dead before the first bet is placed, which is why the gentlest system on the menu still walks the same road to the same loss as the most reckless one.
If you want an approach that actually fits the maths, the honest answer is that no staking pattern improves your odds, so the rational baseline is to stake flat, keep the stakes small and treat the entertainment as the only thing you are buying. Our beginner’s strategy guide sets out what little genuinely helps, none of which involves a progression.