D'Alembert strategy: one-unit steps and why the edge still wins

Jean le Rond d’Alembert’s name is attached to an 18th-century idea that errors cancel: after a miss you were “due” a hit, so you add a unit; after a hit you remove a unit. Modern probability discarded the “due” part. The staking rule survived as folklore.

The operational d'alembert strategy is only the ladder:

1. Choose a unit, say $1, and a starting stake of one unit.
2. If the even-money bet loses, next stake is one unit higher.
3. If it wins, next stake is one unit lower, but not below one unit.
4. Repeat until you stop for a budget or a time limit.

It is meant for 2x bets. It is not meant for Green, parlays or jackpot shares. The Games and odds topic collects the other progressions. Compare the climb with Martingale (doubling) and Paroli (raise on wins). PVPspinArena is 18+.

The name does extra work. An Enlightenment mathematician on the label makes a one-unit ladder sound like a theorem. It is a schedule. Schedules do not move p. If the appeal is “I will not double”, you can not-double with a flat unit and skip the climb that still leaves a hole after the next win.

Base unit $1, Purple-style 2x bet. Start at $1.

Sequence A — mild chop

Five bets, three wins, two losses, +$3. This is the brochure. Notice you needed the $2 win to climb out of the first hole. A flat $1 player with the same W/L string would be +$1 (WWWLL is not this string — here it was L, W, W, L, W: three wins, two losses, +$3 versus flat +$1). The extra $2 came from staking more on a win that happened to follow a loss. That is luck of order, not a theorem.

Sequence B — six losses then a win

Stakes: $1, $2, $3, $4, $5, $6, then $7 to try to recover. Losses total $21 before the seventh bet. A $7 win pays $14 back, so you are still −$14 after the “recovery” win. D’Alembert does not promise that one win clears the slate. That is the trade versus Martingale: slower growth, incomplete repair.

If the seventh also loses, you are −$28 and looking at an $8 stake. Linear is not the same as small.

Sequence C — win first, then a slide

Start $1, win (+$1, next $1), lose (0, next $2), lose (−$2, next $3), lose (−$5, next $4), win (−$1, next $3). You have three losses and two wins and you are still minus. The extra units after the first miss did the damage. Flat $1 on the same WLLLW tape is −$1 (two wins, three losses). D’Alembert turned a −$1 tape into a deeper number because the losses arrived in a clump while the stake was climbing. Order matters for the path; it does not matter for p on the next spin.

Because the step is $1, people pick a unit that is already large: $5 or $10, thinking they have avoided Martingale insanity. Ten losses in a row then take 1+2+…+10 = 55 units. At $5 that is $275 down, with an 11-unit ($55) bet still to come that will not even fill the hole.

On Purple, 10 losses in a row have probability (8/15)^10 ≈ 0.17%, about 1 in 580 isolated starts — and you get many starts in a long sitting. On a fair coin it would be about 1 in 1,024. The extra slots make the ugly path more common, which is the opposite of what a “balanced” story claims.

Floor at one unit

The floor matters. After a win at $1 you stay at $1. Winning streaks do not build a reserve the way Paroli tries to. You grind at the base until a loss starts the climb again. The system is almost designed to be at minimum stake when you are winning and above minimum when you are losing. That is the wrong way round if you believed independence, and independence is the truth. A flat unit accepts that truth in the stake. D’Alembert argues with it in the stake and then agrees with it in the result. Save the argument.

Write EV on each stake: −0.0667 × stake on Purple. The d’Alembert path puts larger stakes after losses. Those larger stakes have the same percentage leak. Expected loss on a cycle is the edge times the dollars actually wagered on that cycle, which is larger than the edge times “I only meant to bet one unit”.

A betting system cannot change p or the 2x payout. It can only choose a stake process. Any process that increases exposure after losses concentrates money on the worse half of the sample path. House edge is collected on exposure.

Expected units on a 20-bet sitting

There is no closed cancel identity. A rough sense: if you start at 1 and the walk is an integer random walk with a downward drift (because q > p on Purple), the stake process spends a lot of time above 1. Average stake of 2 to 3 units over 20 bets is ordinary. That is $40 to $60 turned over at a $1 unit, expected leak about $2.70 to $4.00, plus a left tail if the walk goes to 8 or 10. Flat $1 for 20 bets turns over $20 and leaks about $1.33. The strategy costs more because it wagers more, not because the wheel changed.

Labouchere versus d’Alembert

Labouchere (cancellation) uses a list of numbers and bets the sum of the ends; you cross out numbers after wins and add the lost stake after losses. It is another negative progression with a different schedule. It fails for the same reason: the list blows up on a cold run, and each line is still minus-EV. If a page sells “Labouchere vs d’Alembert” as a skill matchup, both sides are paying the wheel. Fibonacci is the same family with a slower, named sequence.

Coloured wheel

Purple or Silver: 7/15, 2x, 6.67% edge. Rounds are fast, so a linear climb can still eat a stack in a single sitting. Watch the tempo on Roulette before you decide a unit.

European and American

Red/black inherits 2.70% or 5.26%. The ladder is unchanged. The leak per dollar is smaller on European, larger on American. D’Alembert does not close the double-zero.

Coinflip

A 0% fee Coinflip is ~0-EV. The strategy then “fails” in a different sense: it does not create profit, it only correlates larger bets with recent losses. Average P/L tends to zero; the path is choppier than flat betting. You also need an opponent to match the new unit each time.

Jackpot

Do not run d’Alembert on Jackpot tickets. Hit rate is your share of the pot, not ~50%. Adding a unit after a miss is just buying a bigger long shot while tilted.

None of the four rewrites p. Pick among them only if you are choosing a *distribution shape* you already understand is minus-EV. The grown-up choice is flat units and a stop.

If you still want a d’Alembert-shaped night for folklore reasons, cap the pointer: never go above 3 units, and treat hitting 3 as a session warning, not as “the system is working”. That cap is a budget in disguise. Be honest that the disguise is doing the useful work.

If you want the psychological comfort of “I have a rule”, use rules that shrink risk:

• one unit, never increased after a loss;
• a session loss cap in dollars, not in “I’ll stop when the ladder comes back”;
• a time cap, because linear systems invite “just one more step”;
• no borrowed units.

The d'alembert strategy fails because unit size still dies against the edge. A slower death is not a win. If you are already stepping the unit to recover a mood, that is chase. Stop and read the responsible gambling page. Verify past rounds on fairness if you need to see that the Ls were real; then leave the ladder in the 18th century. Gambling is 18+. If you want one number to keep: a linear climb still multiplies the edge by extra units, and one win does not settle the account. Write the running hole in dollars, not in “steps back to one”. When the hole is larger than the session cap, the strategy is over, whether the pointer says 3 or 9. Linear steps feel civilised right up until you add them. Add them on paper first.