Martingale on Stake dice: the honest version

Martingale is the first strategy everyone runs on Stake dice and the one that produces the most 'I was up $400 and then it was gone' stories. Both things are true because martingale is designed to convert many small wins into one large loss. Here's the exact shape of that trade and what the alternatives change.

The rule

Bet at 49.5% (2× payout). Lose → double the bet. Win → reset to base. After any win, your net for the cycle is exactly +1 base bet, no matter how many losses came before it. That is the seductive part: as long as you can keep doubling, every cycle ends in profit.

The doubling table

Why "0.0017%" is not rare

A bot at 4 bets/second makes ~14,000 bets per hour. Roughly half are cycle starts. The expected number of 16-loss streaks per hour is about 7,000 × 0.0017% ≈ 0.12 — one every 8 hours of botting. A 14-loss streak: one every 2 hours. If your balance funds 13 doublings, you should expect to bust in an evening.

Now compare what you earn in that time. Each cycle nets +1 base; at ~7,000 cycles/hour that's +7,000 base/hour — while a single 14-loss bust costs 16,383 base. Over 2 hours you gain ~14,000 and lose ~16,400. That −2,400 is, unsurprisingly, about 1% of the turnover. The house edge doesn't care how you arrange the bets. (Full math in house edge, worked out.)

The balance you'd need

To survive a 1-in-10,000 streak at 49.5% you need to fund 14 losses = 16,383× base. With a $100 balance that means a base bet of $0.006 — and you'd still bust every ~2 hours of continuous play. For a base of $0.10 you'd want ~$1,640 and still expect the same bust rate. There's no base bet small enough to make martingale safe; there's only a base small enough that the bust takes longer.

What the variants change

Delayed martingale

Flat-bet through the first 3–6 losses, then start doubling. Peak bet for a given streak length drops sharply (skip 4, then a 14-streak needs 2¹⁰ = 1,024× instead of 8,192×), at the price of each win only recovering part of the streak. You bleed slowly instead of cliff-diving.

Fractional increase

Increase by 50–80% instead of 100% on each loss. Bet growth is slower, so the balance survives longer streaks, but a win no longer covers the full run of losses. Many "safe" scripts are this plus a reset after a partial recovery.

Deep-chance recovery

Move the win chance down to 1–5% and increase the bet by a few percent per loss. The bet curve is nearly flat compared to doubling, and one hit at 20–99× pays back a very long run. The risk shifts from peak bet to dry-run length — at 1% chance a 460-roll miss happens 1 in 100 cycles. This is the family the Deep 1% Recovery strategy on the DiceStrats marketplace belongs to.

Capped martingale ("never negative" style)

Double up to a hard cap (say 5 doublings, 32× base), then reset and eat the loss. Bust risk falls to near zero because the peak bet is fixed, but you've given up the "every cycle ends +1" property — long streaks now cost you. It's the martingale for people who want to grind wager for VIP progress without the cliff.

If you run it anyway

1. Set a max bet — the bet at which the bot stops doubling and resets or halts. Pick it from the balance you can lose.
2. Set a take-profit that's realistic (5–10% of balance) and stop when you hit it. Martingale's profit is front-loaded; the longer you run, the more certain the cliff.
3. Vault as you go. Auto-vaulting every +X keeps the wins where the next streak can't reach them.
4. Rotate seeds if it makes you feel better — it changes nothing statistically, and DiceStrats can do it every N bets.