Stake dice odds and house edge, worked out

You can run any strategy you like on Stake dice, but you're running it inside one fixed equation. Knowing that equation tells you exactly what a strategy can and can't do — and how to size bets so a normal losing streak doesn't end the session.

The payout formula

Stake dice rolls a number from 0.00 to 99.99. You choose a win chance p (as a percentage) and whether to roll under or over. The multiplier is:

multiplier = 99 / p

The fair multiplier would be 100 / p. Stake pays 99 / p. The difference is exactly 1% at every chance — the house edge is the same whether you bet at 95% or 1%. What changes is the variance: low chance means rare, large wins; high chance means frequent, tiny wins.

What 1% costs per hour

Expected loss = 1% × total wagered. The strategy doesn't matter for this line; only turnover does.

This is why bots feel like they "leak" even when the strategy seems fine — a bot's only superpower is turnover, and turnover is what the edge is charged on. It's also why rakeback matters more to a bot user than to anyone else: it refunds a slice of exactly this number.

Streak probabilities

The probability of losing n times in a row is (1 − p)^n. Two things people get wrong: streaks are far more common than intuition says, and a bot rolls enough times per hour to find them.

Read the last column as "the streak you must be able to survive for a session of ~10,000 cycles". A 49.5% martingale needs to fund 14 doublings — 16,384× base — to survive a 1-in-10,000 event. At 1% chance a recovery strategy needs to fund ~916 consecutive losses. Whichever family you run, size the base bet from that column, not from the average.

"Due" numbers and hot seeds

Each roll on Stake dice is generated from a server seed, your client seed and a nonce, hashed with HMAC-SHA256. The result of roll 1,000 is not influenced by rolls 1–999. A number is never "due", a seed is never "hot", and a streak of 20 losses does not make a win more likely on roll 21 — the probability is still p. Rotating your client seed changes the sequence you get, not its distribution; it's fine as a ritual, but it has no effect on expectation. (Details in how Stake dice is provably fair.)

What strategies actually control

• Shape of the profit curve — steady grind with rare cliffs (martingale, recovery) vs. noisy drift (flat) vs. bursts (Paroli).
• Bust probability per session — determined by peak bet vs. balance.
• Turnover per hour — and therefore expected loss and rakeback earned.

They do not control the sign of the expectation. That's fixed at −1% of turnover. Once you internalise that, strategy choice becomes what it should be: picking the session shape you enjoy and capping the downside with stop-loss and vault. See the strategy families for which shape is which.