The math

Everything here is checkable. Please check it.

This page exists because the rest of this industry depends on you not doing this arithmetic. It takes about four minutes.

Where the edge comes from

A missing pocket

A single-zero wheel has 37 pockets. A straight-up bet on one number pays 35 to 1 — you get your chip back plus 35 more, so 36 units come back on a win.

You win one time in 37. You are paid as though you won one time in 36. That gap is the entire house edge, and it is the only thing the casino needs.

EV per unit staked
= (1/37 × 36) − 1
= 36/37 − 1
= −1/37 = −2.7027%

Pays as if

Chances
1 in 36
35:1 plus your stake

Actually is

Chances
1 in 37
the zero is the difference

Bet selection

No bet on the table is better than any other

Players spend years hunting for the “best” bet. There isn't one. The payouts are constructed so that every wager on a single-zero table returns exactly the same rate. Covering more numbers lowers your variance, never your loss rate.

BetPocketsPaysEdge
Straight up135:1−2.7027%
Split217:1−2.7027%
Street311:1−2.7027%
Corner48:1−2.7027%
Six line65:1−2.7027%
Dozen / column122:1−2.7027%
Red / black, odd / even181:1−2.7027%

Progressions

A martingale doesn't change the rate. It changes the shape.

Take the most common system there is. Bet $10 on red. Double after every loss. Stop-loss of $150, which covers four steps: 10, 20, 40, 80. Here is that system, fully priced.

Most of the time

You win $10
93.05%
feels like a system that works

The rest of the time

You lose $150
6.95%
(19/37)⁴ — four reds missed in a row

Net

Expected per cycle
−$1.13
every cycle, forever

The part that matters

Across one cycle you wager $41.65 on average — sometimes $10, sometimes the full $150. Divide the expected loss by the expected amount wagered:

−$1.1257 ÷ $41.6510 = −2.70270%

The same number. Not close to it — it. The progression moved money between outcomes and left the rate untouched. Every system does this, because arranging losing bets in a clever order cannot produce a winning bet.

What a progression genuinely changes is variance and risk of ruin. A martingale buys many small wins by selling one large loss. That trade is real, it is sometimes what a player wants, and SpinEdge will size it honestly for you — but it is a trade, not an edge.

What automation can honestly fix

Not the odds. Everything else.

If the rate is fixed at −2.703%, the only thing left to improve is how much worse than −2.703% a human does. That gap is large, and it is entirely made of execution mistakes.

Stakes sized wrong

A stake that can't be built from the chips on the table gets silently rounded down, and every number you track afterwards is wrong. SpinEdge decomposes into real denominations before the bet commits, and reports the shortfall when one exists.

Stop losses ignored

Human stop losses lose arguments at 2am. SpinEdge checks whether the next stake would breach the floor before placing it, and refuses the bet rather than discovering the breach afterwards.

Chasing after a loss

The progression is fixed in advance and advances only on a settled, confirmed result. An unacknowledged placement advances nothing.

Not knowing the odds you took

Before you play, the same engine replays your configuration over real recorded spins and reports the expected loss, drawdown, and probability of reaching your target before your floor.

The one real edge

A physically biased wheel — and the four things it takes to prove one

An edge in roulette exists in exactly one situation: a physical wheel with a manufacturing or wear defect that makes some pockets land more often than 1 in 37. A random number generator cannot be biased this way, so this applies only to real wheels with real dealers. It is rare, and casinos maintain wheels specifically to remove it.

  1. 01

    Thousands of spins on one wheel

    Not thousands across many wheels. A bias belongs to a single physical wheel, and pooling wheels manufactures patterns that belong to none of them.

  2. 02

    Correction for everything you tested

    Watch 500 wheels and check every pocket on each, and ordinary luck will hand you 'significant' results. Without a population-wide correction, a fleet-scale scan is a false-positive machine.

  3. 03

    Confirmation after the flag

    The bias must hold on spins recorded after you claimed it. Evidence found and confirmed in the same data is not confirmation.

  4. 04

    An edge that survives the house edge

    A pocket must clear 2.778%, not 2.703% — break-even on a 35:1 payout. Beating fair isn't enough; it has to beat the payout.

Until all four hold, the honest position is that there is no edge, and SpinEdge sits out. Its bias detector is built to stay silent on fair wheels — a tool that finds an edge everywhere has found nothing.

Our own data

Our own data, and what it doesn't prove

0321519421225173462713361130823105241633120143192218297281235326
Deviation from 2.703% · wheel orderfair sharehottest 5 · 2.971%   coldest 35 · 2.594%

A sample of 66,838 spins from live play, in physical wheel order. The chi-square across all 37 pockets is 55.91 on 36 degrees of freedom, giving p ≈ 0.018.

By the usual convention that is “statistically significant”. It is not evidence of a beatable wheel, and the reasons are the first two requirements on the list above: this corpus pools many wheels, and we tested every pocket before picking the interesting one.

We publish it anyway, because a vendor who only shows you the flattering half of their data has told you something about themselves.

Fair share

Per pocket
2.703%

Break-even

Straight up
2.778%
35:1 payout

Hottest · pocket 5

Observed
2.971%
1,986 hits · 4.3σ

Coldest · pocket 35

Observed
2.594%
1,734 hits

Snapshot exported 2026-08-06. Pooled across tables and sessions, and superseded as more play is recorded — the sample size is quoted because a chi-square without one means nothing, not as a measure of how much data exists. methodology