Kelly Criterion for Dice

Kelly answers a precise question: if you can make the same favorable bet repeatedly, what fixed fraction of your bankroll should each bet be to grow your money fastest without risking ruin? The answer balances two forces. Bet too little and you leave growth unclaimed; bet too much and volatility drags your compound growth negative even with the edge in your favor.

For dice the inputs are refreshingly concrete. At 49.5% win chance the payout is 2.00x, so net odds b = 1 (you win one unit per unit staked). p = 0.495, q = 0.505:

f* = (1 x 0.495 - 0.505) / 1 = -0.010

The numerator, bp minus q, is just the expected value per unit staked. That is the quiet elegance of the formula: Kelly is your edge divided by your odds. Positive edge, positive fraction. Zero edge, zero fraction. Negative edge, negative fraction, and since you cannot bet a negative amount, the instruction is: do not bet.

Note what Kelly is not: it is not a progression. It never reacts to the previous roll, never chases, never anticipates streaks. It looks only at the probabilities and odds of the next bet, which for provably fair dice are printed on the interface and verifiable down to the seed hash.

Here is the formula evaluated across the dice slider, using payout = 99 / chance:

Every row is negative, so every row means the same thing: the mathematically optimal wager for bankroll growth is zero. (The f* magnitudes differ because the formula divides the same roughly 1% edge by different odds; they are all just restatements of "no".)

This is the most honest page on our strategy hub because the strategy it describes refuses to play. Martingale, Fibonacci, and Labouchere all answer how to bet into a negative edge; Kelly answers whether, and the answer is no. No staking scheme changes the expected value of a dice roll, and Kelly is the formal proof: when the edge is negative, the growth optimal exposure is none. Everything you wager on a 1% edge game is entertainment spend, and honest sizing starts from that admission.

Kelly becomes a working tool the moment something external tilts the math. Suppose a promotion effectively refunds enough of your play that your all-in expected return on a 49.5% roll rises from 49.5% to 50.1% win equivalent, an effective p of 0.501 at even odds:

f* = (0.501 - 0.499) / 1 = 0.002, or 0.2% of bankroll per roll

A 1 BTC bankroll should stake 0.002 BTC per roll, not more, even though the edge is real. Push effective p to 0.5025 and Kelly says 0.5%. These numbers surprise people: genuine edges in gambling promotions are thin, and Kelly sized bets on thin edges are small. If your instinct on finding a +0.2% edge is to bet 5% of your roll, Kelly is telling you that instinct converts a positive expectation into likely ruin through variance drag.

Where do real cases come from? Rakeback programs that return a fixed share of house edge on every bet effectively reduce the edge (a 20% rakeback on dice turns minus 1% into roughly minus 0.8%: better, still negative, still bet zero by Kelly). Deposit bonuses can go further: if bonus value exceeds the expected wagering cost, the combined package is genuinely positive, and Kelly sizes the play. Sites structure rakeback very differently; our Gamdom dice review works through one of the more generous instant rakeback schemes and what it does, and does not do, to effective edge.

Full Kelly has a brutal property: betting double the Kelly fraction produces zero long run growth, and anything beyond that produces negative growth even on a positive edge game. The penalty for overbetting is worse than the penalty for underbetting, and your edge estimate is always uncertain. If you think your bonus play is worth +0.4% but it is actually +0.2%, full Kelly on the wrong number has you betting exactly at the zero growth threshold of the true edge.

The standard professional response:

• Half Kelly: about 75% of the theoretical growth rate with half the volatility and a wide cushion against edge misestimation. The default for most serious advantage players.
• Quarter Kelly: about 44% of theoretical growth at a quarter of the volatility; appropriate when the edge estimate is soft, which for casino promotions it usually is.

Fractional Kelly also tames drawdowns. Full Kelly bettors should expect their bankroll to halve at some point with probability of roughly one half; that is intrinsic to the growth optimal path, not bad luck. Half Kelly cuts that halving probability to about one in eight. For a dice player working a thin promotional edge, the practical prescription is almost always quarter to half Kelly on an already small f*, which lands on stakes near 0.05% to 0.25% of bankroll, numbers that look a lot like the flat betting guidance in our flat betting guide.

Even when you are knowingly paying the 1% edge for entertainment, the machinery behind Kelly still describes your session. Log wealth math says volatility drag scales with the square of your bet fraction: betting 5% of bankroll per roll does not feel five times swingier than 1%, it produces twenty five times the drag on your compound bankroll path. This is why two players with identical luck and identical total wagered can end a night at minus 8% and minus 60% of bankroll respectively.

Three Kelly derived rules of thumb worth stealing for negative edge play:

1. Bet fractions, never fixed amounts you cannot recompute. Sizing to current bankroll, the core Kelly habit, means losing streaks automatically shrink your stakes and cannot ruin you outright.
2. Halve your instinct. If uncertainty argues for half Kelly when the edge is positive, it argues at least as strongly for conservatism when the edge is against you.
3. Judge systems by growth, not by hit rate. Kelly evaluates strategies on the compound path of the bankroll. By that metric every dice progression scores negative, and the differences between them are just the shape of the decline.

For turning these fractions into concrete unit sizes, session budgets, and stop placements, the companion page is our dice bankroll management guide; Kelly supplies the theory, that page supplies the checklist.