Why the Better Player Still Loses Three Games in Ten — Variance in Backgammon

A concrete walk-through of variance in backgammon: how often the better player actually wins a single game, how match length amplifies skill, and what the math says about losing streaks before you blame the dice.

If you have played backgammon online for more than a week, you have had this thought: the dice are rigged.

Five gammons in a row. The opponent rolls double-sixes off the bar against your five-point board. You have an 80% race lead and lose to four straight doublets. You drop the cube on a position the bots later say was a take, and your opponent rolls bear-off-perfect numbers all the way home. Of course the dice are rigged.

Almost certainly, they are not. What you are feeling is variance — the ordinary, mathematically necessary spread of outcomes around an expected result. Backgammon has more variance than almost any game its serious players claim is a "skill" game, and the disconnect between how skilled the strong players are and how often they still lose is the single most disorienting thing about the game for anyone coming to it from chess, poker, or any other competition that feels less random.

This post is an attempt to name the math behind that feeling.

The headline numbers

Take a 60/40 player. Skilled enough that against a typical club opponent, they would expect to win 60% of single games. (To put that on a scale: this is not a small edge. A 60/40 win rate is a several-hundred-point ELO gap — clearly the stronger player.)

Now: how often does the 60/40 favorite actually win?

A single game is a coin flip with thumb on the scale. Two players in three is the favorite — but one player in three is the underdog, and the underdog wins. A 25-point match brings it close to certainty, but you still drop one match in fifteen against a markedly weaker player, and the math does not care that you were better that night.

Now run the same table for a more realistic opponent — a 55/45 advantage:

A 55/45 player who plays a hundred 7-pointers will go something like 65 wins, 35 losses. Most of the world is in this regime. Most of the time, the better player loses one match in three.

This is not a rigging problem. This is what the game is.

Why backgammon variance is so heavy

Three things stack up.

First, every turn is a roll. Chess has variance only from your own mistakes — a perfect player against a perfect player has zero variance. Backgammon has variance even between two perfect players, because the rolls are random. A perfect player who rolls 1-2 in the opening loses equity instantly to a perfect player who rolls 6-5. The dice make decisions both of you have to live with.

Second, the spread per roll is huge. Each turn is one of 21 distinct rolls (15 non-doubles plus 6 doubles), and the equity differences between them are not small. A double-six on the right turn is worth maybe 15% game-winning equity all by itself. A 1-2 in some positions is worth less than nothing — it loses you tempo and exposes blots. There is no chess move that swings 15% on its own.

Third, the cube multiplies it. A doubling cube does not change the probability you win the game; it changes the stakes of each game. A 60/40 player who plays cubeless wins money at +20% per game. The same player playing optimal cube wins money faster — but each game is now louder. A losing game with the cube on 4 costs four times what an undoubled game costs. Variance scales with stakes.

These three forces — random rolls, big spread per roll, and the cube — combine to give backgammon a variance-to-skill ratio that surprises everyone the first time they take it seriously. Strong play wins eventually. Eventually is longer than most people's patience.

What this means for losing streaks

The math also tells you what to expect when you are losing.

A 55/45 player has roughly a 6% chance of losing any 5 specific matches in a row. Over a hundred matches, you will see a five-loss streak more often than not — about 60% of players will hit one somewhere in the run. A seven-loss streak comes up around 1% of the time per starting point but, again, over a hundred matches the chance you encounter one is meaningful.

If you sit down and play 100 games of backgammon as a 55/45 player, the math says:

• You will win 53–57 of them. (One standard deviation. The 95% range is roughly 45 to 65.)
• You will hit at least one stretch of 5 losses in a row.
• You will hit at least one stretch of 4 wins in a row.
• You will, with probability around 1 in 6, finish below 50% even though you are the better player.

The dice are not rigged. The math is.

The cube is a variance amplifier (and a skill amplifier)

Here is where it gets interesting. In a peer-to-peer money game, you can choose how much variance you take on by how aggressively you play the cube. A scared cube — never doubling unless you are at 80% — bleeds equity to your opponent. An aggressive cube — doubling at the proper take points — extracts maximum equity but turns every game into a louder swing.

The right move is the aggressive cube, and a long enough sample to tell skill from noise. The wrong move is to play short matches against people you know are weaker than you and decide on game 30 that you are running bad.

The doubling cube post goes deep on the actual cube mechanics. The point here is narrower: when you understand the cube, you understand that bigger games and bigger swings and more skill expression are the same thing. You cannot have one without the others.

"But what if it really is rigged?"

You have read this far and you might still suspect the dice. Online platforms have, historically, given players reasons to be suspicious — ranging from sloppy RNG implementations to outright cheating in a few documented cases. The suspicion is rational; the question is what you can actually do about it.

On 6proclub, the answer is: re-derive every roll yourself.

Every dice roll on the platform is generated by a published commit-reveal protocol. Before each game, the server commits to a random seed by publishing SHA-256(seed) — a one-way hash. Your client contributes its own random seed too. Each roll is derived from SHA-512(server-seed : client-seed : turn : roll-index), and the server reveals its seed at the end of the game. You can re-compute every roll yourself.

We ship a tool for it. Open the roll verifier, paste in any finished game's seeds and commits, and the tool will re-derive the dice in your browser using crypto.subtle.digest() — the same primitive your bank uses. If the recomputed dice match the dice you played, the rolls were untouched. If they don't match, something is wrong. So far they always match — and if they ever don't, you can prove it without our cooperation.

Most of the time you suspect the dice, the math is responsible. But you do not have to take that on faith.

The strategic takeaway

If you remember three things:

1. Variance is heavy in backgammon. The better player loses 30–40% of single games, which feels rigged but is just dice.
2. Match length is signal length. A 25-point match is reliable. A single game is not. Build your rating, your money game, and your impression of an opponent over many matches, not one.
3. The cube is the price of skill expression. It also amplifies short-run swings. Take the swings, take the upside, and trust the longer sample.

Lose, then play another. Lose again, then play another. After enough games, the math wins. That is the only deal backgammon offers, and it is the only deal worth taking.

→ Want to verify your last game's dice? Open the roll verifier.