Every roulette system in existence makes the same quiet promise: that a betting pattern can extract profit from a wheel whose outcomes are random. The promise has survived three centuries, countless bankruptcies and the complete mathematical analysis of the game, which suggests the systems sell something other than profit — and indeed they do. They sell structure. A system tells you what to bet next, which replaces the anxiety of choosing with the comfort of a rule, and for many players that comfort is worth the price even knowing the price exists. But the price varies enormously between systems, and that variation is where the real analysis lives. Martingale, D’Alembert and Fibonacci are the three most common progression systems, and they differ from each other in ways that matter: how fast the stakes grow during a losing streak, how much bankroll each requires to survive a given losing run, and how quickly each one recovers losses when the wheel finally cooperates. This guide runs all three through the actual numbers — streak probabilities on real wheels, bankroll requirements, expected values and the practical failure points that every one of them shares — because the interesting question is not whether systems beat roulette (they do not, and the proof is short) but which system survives longest with a given budget, and what each one actually costs in exchange for the structure it provides.
The one fact that governs everything
Before any system can be evaluated, one property of roulette must be stated precisely, because it is the property every system fails to overcome. European roulette has 37 pockets; a bet on red, or any even-money bet, wins with probability 18/37 and loses with probability 19/37. The expected value of any single unit wagered on an even-money bet is therefore:
Every even-money bet loses an average of 2.7 cents per euro, no matter how it is sized, sequenced or motivated. This is the house edge, and it applies to every bet on the layout except the five-number American bet, which is worse. The critical consequence for betting systems is this: a system is a sequence of bets, and the expected value of any sequence of negative-expectation bets is the sum of the expectations of those bets — always negative, regardless of the pattern. No sizing rule, no stop-loss, no progression changes the sign of the expectation, because the expectation is a property of each individual bet, not of the order in which they are placed.
What a system can change is the distribution of outcomes: it can trade many small wins for occasional catastrophic losses, or spread the pain more evenly, or stretch the bankroll across more spins. Changing the distribution without changing the expectation is the honest description of every system ever sold — and the distributions differ enough between the three systems that the comparison is genuinely worth making.
Martingale: the double-or-nothing classic
Martingale is the simplest and most dangerous progression ever devised: bet one unit on an even-money chance; if it loses, double the stake; continue doubling after every loss; when a win finally arrives, the net profit equals the original stake. The system’s logic is airtight at the limit — an eventual win always recovers all losses plus one unit — and that airtightness is precisely the illusion, because the limit requires an infinite bankroll and a table with no maximum.
The reality is bounded on both ends, and the numbers show how tight the bounds are. A losing streak of consecutive spins is not a rare anomaly at roulette: the probability of losing 10 even-money bets in a row is (19/37)10≈0.00078, which sounds negligible until you realize a player running Martingale for a single evening of 100 spins faces that streak with roughly 7.5% probability — once in every thirteen sessions or so. The stakes tell the rest of the story.
| Losing streak length | Stake required on the next bet | Total amount lost so far | Probability of the streak occurring in 100 spins |
|---|---|---|---|
| 5 in a row | 32 units | 31 units | ~48% |
| 7 in a row | 128 units | 127 units | ~16% |
| 10 in a row | 1,024 units | 1,023 units | ~7.5% |
| 12 in a row | 4,096 units | 4,095 units | ~2.7% |
| 15 in a row | 32,768 units | 32,767 units | ~0.55% |
The table’s brutal message is the asymmetry between reward and risk. A Martingale player risking a 1,024-unit streak to win 1 unit per cycle is operating at a risk-to-reward ratio of a thousand to one, and the table shows the streaks arrive with uncomfortable regularity. Meanwhile, most roulette tables cap even-money bets at levels reached within 8 to 12 doublings of the minimum — the table maximum is not a theoretical obstacle but a rule printed on the felt, and it ends the progression before the bankroll does. When either limit is hit, the entire accumulated loss of the streak is realized at once, wiping out the profits of dozens of successful cycles.
The system’s defenders offer a variant — starting with a higher unit so the profit per cycle justifies the risk — but this changes nothing except the scale of the catastrophe. Martingale does not fail because the player runs out of patience; it fails because doubling is exponential, and exponential growth always outruns finite bankrolls and table limits before the mathematics of the wheel can cooperate.
D’Alembert: the gentler climb
D’Alembert was designed as a correction to Martingale’s madness, and it works by replacing multiplication with addition. The rule: bet one unit; after each loss, increase the stake by one unit; after each win, decrease it by one unit. The progression climbs linearly rather than exponentially, which transforms the entire risk profile.
The arithmetic shows why. After a ten-loss streak, a D’Alembert player has staked 1+2+3+…+10 = 55 units and faces a next stake of 11 — against Martingale’s 1,024. The system survives streaks that annihilate Martingale completely, and a single win during recovery pulls the stake down the ladder quickly. The recovery logic is also elegant: because the stake is one unit higher after each loss than the one before it, a win at any point in the sequence recovers the loss of the previous bet and returns the progression toward equilibrium.
The trade-off is equally clear: recovery is slow. After the same ten-loss streak, Martingale’s next win erases everything at once; D’Alembert must grind through a sequence of wins, each at declining stakes, to return to break-even — and the wheel owes no such sequence. The system’s expected value is identical to Martingale’s: negative, fixed, 2.7% per unit wagered. What D’Alembert changes is the shape of the session: far lower ruin risk, far slower profit accumulation, and occasional long stretches where the stake sits above the starting unit after a run of losses, bleeding the bankroll at an elevated rate.
Fibonacci: the sequence with a memory
The Fibonacci system borrows the famous sequence — 1, 1, 2, 3, 5, 8, 13, 21, 34, 55 — where each number is the sum of the two before it. The rule: move one step forward in the sequence after each loss, two steps back after each win. The system sits deliberately between Martingale and D’Alembert in aggression: its stakes grow faster than linear addition but far slower than doubling.
After the same ten-loss streak, the Fibonacci player has staked 1+1+2+3+5+8+13+21+34+55 = 143 units and faces a next stake of 89. Compare the three systems at the same point: Martingale has lost 1,023 and needs 1,024 more; Fibonacci has lost 143 and needs 89; D’Alembert has lost 55 and needs 11. The Fibonacci two-steps-back rule, however, gives it a recovery property the others lack: because each win retreats two positions while each loss advances one, the sequence returns to its start quickly once wins arrive — a Fibonacci player needs wins on roughly one-third of the recovery bets to climb back to the beginning, whereas D’Alembert needs a win on roughly half.
The comparison across all three systems, evaluated on the same criteria:
| Criterion | Martingale | D’Alembert | Fibonacci |
|---|---|---|---|
| Stake growth pattern | Doubles every loss | +1 unit per loss | Fibonacci step per loss |
| Total loss after 10-loss streak | 1,023 units | 55 units | 143 units |
| Next stake after the streak | 1,024 | 11 | 89 |
| Recovery after one win | Full recovery | Partial (one step down) | Two steps down |
| Realistic ruin risk | Extreme | Low | Moderate |
| Time to recover a losing streak | One winning spin | Many spins, slow | Moderate, sequence-based |
| Expected value per unit | −2.7% | −2.7% | −2.7% |
The last row is the punchline the entire analysis has been building toward: all three systems are mathematically identical in the only dimension that ultimately matters, and they differ only in how the identical expected loss is distributed across time. Martingale concentrates the loss into rare, devastating events; D’Alembert spreads it thinly across every session; Fibonacci splits the difference. Choosing between them is choosing a payment plan for the same bill.
What the math actually proves
The analysis can now be compressed into findings that are short enough to state and firm enough to rely on:
- No progression changes the expected value. The house edge of 2.7% on European even-money bets applies to every bet individually, and a sequence of such bets sums to negative expectation in every pattern — this is arithmetic, not opinion, and it holds for Martingale, D’Alembert, Fibonacci and every unlisted relative.
- Systems change ruin probability, not profitability. A gentle system survives longer and loses more slowly; an aggressive system survives shorter and loses faster. Neither wins in expectation, but their failure timelines differ dramatically — and for a player whose primary constraint is a fixed entertainment budget, the difference in survival time is the entire practical distinction.
- Table limits are a structural counterargument. The house prints maximum bets precisely because progression systems require unbounded stakes; the maximum ends any progression at the point where it becomes ruinous, converting the theoretical guarantee of eventual recovery into an impossibility.
- The win-streak illusion is real and documented. System players experience many profitable sessions and rare catastrophic ones, and human memory weights the frequent wins heavily — which is why the systems survive centuries of refutation: their failure mode is statistically invisible until it happens.
The honest summary of these findings is not “systems are worthless” but “systems are restructuring tools.” A player who wants the session to feel orderly, who wants rules instead of decisions, gets genuine value from a progression system — the same value a budget gets from any structure — and the mathematics permits that value as long as the expectation is understood honestly.
If a system is still going to be used: the safety configuration
Given that the three systems are a choice of loss-distribution rather than a choice of profit, the rational way to use one is as a session framework with hard boundaries. The configuration that extracts the structure while capping the damage:
- Set the unit at 0.5–1% of the session bankroll before the first spin, and treat it as fixed for the entire session — the unit is the foundation of every subsequent stake in all three systems, and an inflated unit compounds through every progression step.
- Choose D’Alembert or Fibonacci for anything beyond casual play. Martingale’s ruin probability makes it unsuitable for bankrolls that matter; the gentler systems deliver most of the same structure at a fraction of the tail risk.
- Cap the maximum stake in advance — for example, ten times the unit — so the progression has a legal stopping point that the bankroll, not the table limit, defines.
- End the session on the pre-committed loss limit, because the moment a system’s player starts breaking the system — skipping steps, restarting the sequence early, chasing outside the rules — the structure’s value evaporates and only its costs remain.
- Log the sessions. The running record converts the win-streak illusion into data, and across ten or twenty sessions the negative drift becomes visible in the player’s own numbers rather than in a textbook’s.
Conclusion
The mathematics of roulette systems is settled, and it says something more interesting than the usual dismissal: Martingale, D’Alembert and Fibonacci are not three roads to profit but three different payment schedules for the same guaranteed cost — the 2.7% house edge, applied to every unit wagered, in every pattern, forever. Martingale pays the bill in rare catastrophes, D’Alembert in installments, Fibonacci in medium-sized payments with a clever recovery mechanism that redistributes the timing but not the total.
The rational use of a system, then, is neither the hope of beating the wheel nor the mockery of those who try. It is the deliberate purchase of structure: a rule that decides the next stake, a session that ends on a pre-committed limit, and a risk profile matched to the bankroll that funds it. D’Alembert with a 1% unit and a hard stop is a sustainable entertainment framework; Martingale with any bankroll short of an institution’s is a countdown to a single bad evening. The wheel does not care which is chosen — the numbers above already did, and they will deliver the same verdict to every player who runs the test, one spin at a time.