The D’Alembert roulette strategy increases an even-money wager by one unit after a loss and decreases it by one after a win. Compared with Martingale, its growth feels moderate. That makes the drawdown easier to overlook, not impossible.
Rules
- Choose an even-money bet and base unit.
- Start with one unit.
- Add one unit after each loss.
- Subtract one unit after each win, never betting below one.
- Stop at a predetermined loss, time or spin count.
Reverse D’Alembert switches the movements: add after wins and subtract after losses. The two versions create different paths but operate on the same negative-expectation bets.
Example sequence
For results L, L, W, L, W, W, stakes become 1, 2, 3, 2, 3 and 2. Net result is −1 −2 +3 −2 +3 +2 = +3 units. The example works because the larger stakes happen to coincide with several wins. Reorder the same outcomes and the result can change.
Linear growth
After uninterrupted losses, stakes follow 1, 2, 3, 4 and so on. The total required for n losses is n(n+1)/2 units. Ten losses consume 55 units and the eleventh stake is 11. Twenty losses consume 210 units.
This is much slower than Martingale’s exponential total of 1,023 units after ten levels. The comparison does not make D’Alembert profitable; it means the system exchanges rapid recovery attempts for a longer, slower progression.
The balancing intuition
The system is often justified by assuming wins and losses eventually balance. Even if counts become similar, their order matters because stakes differ. Roulette even-money decisions are also not balanced exactly: zero causes a European bet to lose 19 times out of 37, and an American bet to lose 20 times out of 38.
Probability does not actively correct a short-term imbalance. A run of losses does not cause the next spin to favour the chosen side.
Recovery behaviour
A win at a larger stake can recover several earlier losses. But because the stake decreases after that win, alternating outcomes do not guarantee a steady climb. A session can remain below its starting bankroll while the sequence moves up and down for many decisions.
Risk from long sessions
Linear progressions are often played longer because individual increases feel manageable. More spins increase total action. At a fixed negative expected value, additional action increases expected monetary cost even when no single stake looks extreme.
Table maximum and bankroll
D’Alembert reaches a table maximum slowly, but a maximum still exists. More commonly, a stop-loss ends the sequence before theoretical recovery. Any performance report must count those incomplete cycles rather than assuming play continues until profit.
European versus American roulette
Because the strategy normally uses even-money bets, La Partage or En Prison on a French single-zero table can reduce the edge materially. Selecting better rules has a direct mathematical effect. Changing from D’Alembert to another progression does not.
Testing protocol
Specify base unit, maximum stake, bankroll, session length and reset conditions. Run identical outcome sequences through D’Alembert and flat betting. Using the same sequences isolates the effect of stake sizing from random luck.
Report total amount wagered in addition to net result. A strategy can finish with a similar loss while cycling far more money through the table.
Advantages and limitations
| Practical feature | What it means |
|---|---|
| Simple arithmetic | Easy to follow and automate |
| Linear increases | Slower stake growth than doubling |
| Path dependent | Order of wins and losses matters |
| No predictive input | Does not choose a more likely outcome |
| Negative expectation | House edge remains on every unit wagered |
Verdict
D’Alembert is less aggressive than Martingale but rests on no predictive advantage. It can be a useful teaching model for linear progressions and path dependency. Before considering any financial use, calculate the full triangular progression with our bankroll tool and compare it with other roulette strategies.
D’Alembert frequently asked questions
Does the law of equilibrium support the system?
Random sequences can show roughly balanced counts over large samples, but deviations do not correct on demand. More importantly, zero makes the chosen even-money side lose slightly more often than it wins. Similar counts do not guarantee profit when stakes depend on order.
What is Reverse D’Alembert?
It increases one unit after a win and decreases after a loss. That makes it a positive progression seeking to exploit winning runs. It does not change probabilities and should be tested as a separate strategy.
When does the standard system reset?
Many descriptions remain at one unit after a win at the base. Others reset after a profit target. State the rule explicitly; discretionary resets create hindsight bias.
Can a linear progression avoid table limits?
It reaches them more slowly, not never. A long net losing period can still push the required stake beyond a maximum or affordable amount.
Path-dependency exercise
Take ten results containing five wins and five losses. Run them in alternating order, then group all losses first. Counts are identical, yet stake paths and ending results differ. This exercise directly disproves the idea that eventual numerical balance alone determines profit.
Measuring session length
Because the system can drift rather than terminate quickly, compare both a fixed-spin session and a fixed-time session. A live table and rapid RNG game can produce very different decision counts per hour. Total action is the more comparable exposure measure.
Zero and half-loss rules
On a La Partage table, a zero may lose half a stake. Decide whether the next bet increases by a full unit, half unit or not at all. No choice should be inserted after seeing which produces the best backtest. Publish it as part of the method.
Interpreting a low maximum stake
A system can keep its single largest stake modest while accumulating substantial losses through many decisions. Report cumulative action and drawdown. Maximum bet is only one dimension of risk.
Responsible conclusion
D’Alembert’s gradual pace makes it easy to understand and easier to cap than exponential systems. It also makes escalation feel harmless. Predefine the maximum level and duration, and treat any breach as a failed test rather than a reason to rewrite the rules.
How to use this research
Treat the worked examples as explanations, not forecasts. Reproduce the assumptions with a free tool, change one variable at a time and keep losing as well as winning trials. A short favourable run cannot validate a strategy, while a short unfavourable run cannot measure its full distribution. For any real-money decision, verify local law and operator rules, set an affordable limit before play and never use a progression to recover money that has already been lost.