Three simulators on the same bet spread
Before you lean on a simulator’s numbers, it helps to know they agree with others. We ran the same scenario through ours and two established tools, PBS from Blackjack Apprenticeship and blackjack.tools, then compared the results. The scenario is a hard one for all three: a deep shoe at a full table, played two hands. The numbers landed within a couple percent of each other.
The test
One configuration, entered the same way into all three tools. Deep penetration and a full table are both places where simulators can drift apart, since each one changes how often you reach the high counts that carry the bet.
- Game
- 8 decks, H17, DAS, no surrender, split to 4
- Penetration
- 0.75-deck cutoff (~90%)
- Table
- You plus four other players
- Count
- Hi-Lo with expanded H17 deviations
- Bet spread
- $25 unit, 1–12 ($25 → $300)
- Two hands
- From true count +2 upward
- Bankroll / pace
- $15,000 · 80 rounds/hr
- Sample
- 1B rounds
What each one reported
Here is what each simulator reported for the win rate, the hourly swing (one standard deviation), risk of ruin on a $15,000 bankroll, and N₀, the hours of play it takes for your edge to overtake the variance.
| Simulator | Win rate /hr | Swing /hr | Risk of ruin | N₀ |
|---|---|---|---|---|
| AdvantagePlay (ours) | $60.48 | ±$1,052 | 19.4% | 303 h |
| PBS | $61.90 | ±$1,030 | 17.3% | 277 h |
| blackjack.tools | $59.21 | ±$1,010 | 17.5% | 291 h |
Our win rate lands almost exactly between the other two, within about 2% of each, and the standard deviation, the figure that most tightly constrains a Monte Carlo run, agrees to within about 2 to 4%. With a billion rounds behind it our own figure is precise to roughly half a percent, so the small remaining differences are real rather than noise, though still minor. Ours carries a slightly higher standard deviation, the cautious direction, which keeps our risk of ruin and N₀ a little above the other two. Most of the rest is how finely each tool models the deviation list and the variance of two hands at once. We ran the same check against Casino Vérité separately; see our CVCX validation.
What this means
Three independently built engines, given the same scenario, returned figures within a couple percent of one another. When our numbers line up with tools people already rely on, the planning figures our simulator produces (win rate, risk of ruin, N₀, and the long-run outcome band) are ones you can size a bankroll around.
Methodology
Our figures are from one billion rounds on the AdvantagePlay engine, a hand-by-hand Monte Carlo that deals every seat from a shared shoe. Eight decks, 0.75-deck cutoff, four other players, Hi-Lo with expanded H17 deviations, H17 / DAS / no surrender / no resplit aces / split to four, a $25 1–12 ramp with two hands from true count +2, a whole-deck true count, 80 rounds/hr and a $15,000bankroll. The PBS and blackjack.tools figures were read directly off each tool configured to the identical game, spread and deck-estimation rule. “Swing” is one standard deviation per hour; N₀ is (swing ÷ win rate)², the hours for edge to overcome variance.