Bankroll Sizing at a 0.22% Edge: The Ruin Curve Nobody Plots

273,000 hands.

That is roughly how long a blackjack player has to sit at a 0.22% house edge before the edge becomes the largest force acting on their balance. Below that, the thing moving the money is variance. Players compare house edge figures to three decimal places and then bring a bankroll sized by feel, which is a bit like tuning an engine and then choosing tires by color.

Where the crossover number comes from

Two inputs. The first is the per-hand expected loss, which at a 0.22% edge is 0.0022 units of your base bet. The second is the per-hand standard deviation, which for a normal rule set with doubles and splits runs a little over one unit; 1.15 is the usual working approximation and it is an approximation, inflated above 1.0 by the hands where you put out extra money.

Expected loss grows with the number of hands. Standard deviation grows with its square root. Set them equal: 0.0022N = 1.15 x sqrt(N), so sqrt(N) = 1.15 / 0.0022 = 522.7, so N = 273,200 hands. At a live table, whatever pace you assume, that is thousands of hours.

Look at the 500-hand row, which is one long evening. The game's designed cost to you is a single betting unit. The routine spread of outcomes is twenty-five. Every session you will ever describe as good or bad was decided by the second column being drowned in the third.

The rebate moves the mean and leaves the noise alone

This is the mechanism people miss. A rebate credited per wager is deterministic: it arrives whether you win or lose, so it shifts the center of the distribution without touching its width.

Take the published figures for Duel Blackjack, where a 99.46% base return becomes a stated 99.78% effective return once a rakeback of up to 60% is credited per wager. Over 5,000 hands, the base edge of 0.54% costs 27 units and the rebated edge of 0.22% costs 11 units. The rebate is worth 16 units. Standard deviation over the same 5,000 hands is 81 units. So a structure that cuts the house edge by more than half moves the distribution by 16 / 81, or about a fifth of one standard deviation.

That is a real improvement and it is the best kind, because it is free of rollover conditions. It is also not a change in risk. The shape of the curve is identical; the whole thing has slid sideways by a fifth of a standard deviation. Anybody who plays longer sessions because the edge got smaller has increased their exposure to the only variable that ends sessions.

Kelly declines to answer

The Kelly criterion sizes a bet as edge divided by variance, and it is worth being blunt about what it says here. The player's edge at this table is negative. Kelly's prescription for a negative edge is a stake of zero, and no rebate below 100% of the house edge changes that: returning 60% of 0.54% leaves 0.216% against you. Only a rebate exceeding the entire house edge would flip the sign, and that product does not exist for a reason.

Run the formula anyway as a scaling exercise. If a player did hold a 0.22% edge, full Kelly would stake 0.0022 / 1.15 squared, which is 0.0022 / 1.3225, or 0.166% of bankroll per hand. To bet ten dollars a hand on that basis you would need about six thousand dollars behind it. That ratio is the honest measure of how thin a fifth of a percent is.

Size from sigma

Since drift is negligible on any human horizon, bankroll is a pure volatility question, and the arithmetic is not hard. Over 5,000 hands the standard deviation is 81 units. A player who wants the probability of ever touching zero to sit near one percent needs a cushion of roughly two and a half standard deviations, which is about 200 units; three standard deviations, about 250 units, buys a little more comfort. Those figures use the reflection approximation for a walk with almost no drift, and they understate ruin slightly because the drift is against you. A hundred-unit bankroll across that same horizon fails somewhere around a quarter of the time.

The uncomfortable part follows immediately. Halving the house edge barely moves that requirement, because the requirement is a multiple of sigma and the edge is not in the formula in any meaningful way. A 0.22% game and a 0.54% game need bankrolls that differ by a rounding error, which is the opposite of what the marketing implies and exactly what the math demands.

Low edge is a discount on the price of playing. It is not insulation, it is not safety, and it does nothing whatsoever for the player who arrives with forty units and a plan to double them. Cheap games break small bankrolls at almost exactly the same rate expensive ones do. They just take marginally longer to do it.