The Simulator is a mathematical tool wearing the clothes of a real table: fourteen rooms, each dealt the way its jurisdiction actually deals it, free, with play money, and nothing to install. Anyone can say their simulator is accurate. This page is the breakdown of what that word means here, item by item, with the way to check each claim yourself - because the whole point of the thing is that nothing about this game needs to be taken on trust.
Every number The Simulator shows you is counted, not sampled - the same standard the book holds itself to.
| What | How it is done | How to check it |
|---|---|---|
| Hand evaluation | Every one of the 22,100 three-card hands scores to a single exact integer - category and kickers together - so comparing two hands is one comparison, with no rounding anywhere. | The scoring is asserted hand family by hand family in the public test suite. |
| Strategy | Not a lookup table. The game is solved live in your browser: 1,755 suit-distinct hand classes, each against all 18,424 dealer hands - the full 407,170,400 confrontations - and re-solved whenever you change tables, because a different pay table is a different game with a different correct play. | Watch the advisor's figures change when you switch rooms; the solve is the same one the reference code prints. |
| Exposed-card rooms | Where the dealer shows a card, your decision is solved fresh against the 1,128 hands he can still finish with, priced for the exact wagers you have out - not marked against the memorised Q-6-4 rule, which is wrong in those rooms. | The printed rule in the book is itself checked against all 22,815 hand-and-upcard positions in the test suite. |
| Two implementations | The browser solver is cross-checked against an independent Python implementation written separately. A figure appears in the book only when the two agree exactly. | Both are public. Run crosscheck from each and compare the six headline
figures to ten decimal places. |
| The money | Settlement is audited by replaying real auto-played sessions and recomputing every hand independently from the raw cards - ante and play, the bonus, Pair Plus - and the two totals match to the dollar, hand by hand. | The audit method is described in the repository; the settle rules are the published ones on each room's sign. |
| The shuffle | A cryptographic random source and a fresh fifty-two card deck every hand. The book's short-riffle model is available as an experiment, and it is labelled as one. | Chapter 35 of the book, and shuffle.py in the repository. |
| The meter | The progressive seeds at $5,000 against a break-even of $7,892.31, so the dollar bet is a bad one most of its life and turns good rarely - which is the lesson the real thing teaches. A permanently generous meter would teach the opposite one. | Appendix E prices the break-even exactly; the ladder on the sign is the one it prices. |
| The drill | Marks every decision against the exact solution for the situation in front of you, with the running cost of your mistakes in antes - not against a rule of thumb. | Play a marginal hand both ways and read the EV it quotes for each. |
The voices, the cocktail service and the pit calling for a verification are furniture - there so that the procedure you rehearse here is the procedure a real table runs, down to the settle order. The furniture is honest too, but it is not the point. The arithmetic is the building.
A tool that claims it was always right is asking for faith. This one keeps its corrections in public.
The solver's original suit-handling merged hands that were not equivalent - 741 classes where the truth is 1,755 - and every ante-and-play figure was slightly wrong until a symmetry check caught it: remove the qualification rule and wins must equal losses exactly, and under the bug they did not. That check now runs before anything else, in public, every time the test suite runs.
And the progressive meter briefly seeded at $60,000 - eight times its own break-even - which quietly made the worst bet on the layout the best one. It was found because sessions kept finishing up when the arithmetic said they should not, and the fix is described above. Both stories are told in full in the book's Appendix H and in the repository history, because a record of corrections is the only accuracy claim that ages well.
The reference code is public - github.com/keithadler/win-at-3-card-poker, MIT licensed, standard library only, written to be read. If you check it and it disagrees with anything here, say so - a correction is more welcome than a compliment.
The advisor, the drill and the session record are all computed live from the arithmetic on this page.
Open The Simulator Read the numbers