Progression systems keep resurfacing because they win most of the time. That is exactly what they are engineered to do, and it is not the same thing as being profitable. We work the Martingale to its exact expected value on a single-zero wheel, then show the one-line theorem that disposes of every system in the family at once — including the ones nobody has invented yet.
Every few months the same question arrives at the desk in slightly different wording: if I double after each loss, surely I must come out ahead eventually? The persistence of the question is not stupidity. It reflects something genuinely true about progression systems — they do win the overwhelming majority of the time. The Martingale player really does walk away up on most nights. The error is in the leap from 'wins most of the time' to 'wins money', and the distance between those two statements is exactly the house edge.
What a betting system actually is
Strip away the branding and every system in the literature is a rule for choosing your next stake based on what happened previously. Martingale doubles after a loss. D'Alembert adds one unit after a loss and subtracts one after a win. Fibonacci walks up the famous sequence. Labouchère crosses numbers off a written line. Oscar's Grind raises only after wins. None of them change which bet you make, only how much you put on it and when.
That distinction is the whole subject. A single-zero roulette wheel has 37 pockets, and an even-money bet on it wins 18 times in 37. Every spin, independently, has the same expectation — you lose 1/37 of whatever is on the table, about 2.70%. The wheel has no memory of your last stake and no knowledge of your staking plan.
The Martingale, worked all the way through
Take the most common real-world configuration: a $10 minimum, a $1,000 maximum, and an even-money bet on a single-zero wheel. The staking ladder runs $10, $20, $40, $80, $160, $320, $640 — seven bets. The eighth would need $1,280 and the table will not accept it. So a cycle either produces a $10 profit or it hits the wall and takes a $1,270 loss.
| Losing seven consecutive spins | (19/37)⁷ ≈ 0.9415% |
|---|---|
| Probability the cycle succeeds | ≈ 99.06% |
| Profit when it succeeds | +$10 |
| Loss when it fails | −$1,270 (10+20+40+80+160+320+640) |
| Expected value per cycle | 0.9906 × (+10) + 0.0094 × (−1,270) = −$2.05 |
So the shape of the thing is exactly as advertised: you win $10 on roughly 99 cycles out of 100. You also lose $1,270 on the hundredth, which erases 127 of those wins. The system is not a way of beating the wheel. It is a way of borrowing many small certain-feeling wins against one large rare loss, and the interest rate on that loan is the house edge.
The cross-check that makes it undeniable
If the Martingale genuinely did anything to the mathematics, its expected loss would differ from the raw edge applied to money staked. It does not. Summing the stake at each rung weighted by the probability of reaching it gives an expected total staked of $75.94 per cycle. Multiply that by the wheel's 2.70% and you get $2.05 — the same figure the cycle calculation produced, to the cent.
The one line that kills the entire family
Here is why nobody needs to test the next system that comes along. Expected value is linear: the expectation of a sum of bets equals the sum of their individual expectations, and this holds whether or not the bets are independent, and whether or not their sizes were chosen based on earlier results. Each even-money spin on a single-zero wheel returns −2.70% of whatever is staked on it. A system chooses the stakes; it cannot touch the −2.70%.
Therefore the total expected result of any staking plan whatsoever is −2.70% multiplied by the total amount it ends up staking. There is no arrangement of positive stakes that makes a sum of negative terms positive. This is not a claim about the systems that exist — it is a claim about every system that could ever exist, including the one being sold in an inbox somewhere this morning.
- Martingale — double after a loss. Rare, enormous losses; frequent tiny wins.
- D'Alembert — up one unit after a loss, down one after a win. Gentler ladder, same expectation.
- Fibonacci — walk the sequence up on losses. Slower escalation than Martingale, identical conclusion.
- Labouchère — cross off a written line of numbers. More bookkeeping, no mathematical difference.
- Oscar's Grind — raise only after wins. Lower variance than the others; the expectation is untouched.
Two things the systems genuinely do
It would be wrong to say progression systems have no effect. They have two, and neither is the one they are sold on. First, they reshape variance: Martingale converts a symmetric spread of outcomes into a heavily skewed one, with a long thin tail of catastrophic sessions. Second — and this is the part that matters outside the mathematics — they systematically increase the amount staked, because escalating after losses is precisely a rule for betting more when you are behind.
That combination is worth naming plainly. A staking plan that raises your bets as losses accumulate, while producing a run of small wins that feels like evidence it is working, is close to a worst-case structure for anyone at risk of gambling harm. The false confirmation arrives constantly; the bill arrives once.
The honest summary of the field is short. The house edge is a property of the bet, fixed by the rules of the game. The only levers a player genuinely has are choosing bets with a smaller edge — the derivations in First Principles price them all — and deciding how much to stake in total. Everything else on offer is a rearrangement.
Does the Martingale work if the table has no maximum?
Only if you also have infinite money and infinite time, which is another way of saying no. With an unlimited table and an unlimited bankroll the strategy does reach a $10 profit with probability one — but the expected amount you must stake to get there is unbounded, and any real bankroll is finite. Every actual table has a maximum, and that maximum exists precisely to cap the ladder.
Is a system with lower variance, like Oscar's Grind, mathematically better?
Better in the sense of a smoother ride, not in the sense of expected return. Lower-variance staking plans produce narrower outcome distributions, so results cluster nearer the expectation. Since the expectation is negative, clustering nearer to it simply means losing at a steadier and more predictable rate. Variance is the shape of the distribution; the edge is where its centre sits.
Why do system sellers show winning results?
Because they are real. A short sample of Martingale sessions will usually show a profit — roughly 99% of individual cycles succeed in the configuration above. Any test small enough to fit in a sales page is far too small to encounter the rare loss that pays for all of it. The results are not fabricated; the sample is just cut off before the bill arrives.
Is there any bet where the expectation is not negative?
Among standard casino table bets, no. The nearest thing is the craps free-odds wager, which pays exactly true odds and carries an expectation of exactly zero — but it may only be placed behind a line bet that does carry an edge, so the combination remains negative. The derivation is in our pass line and free odds study.