Video Poker — The Player’s Own Contribution to the Return, Measured
Full-pay 9/6 Jacks or Better. The paytable is fixed and the return is not: it depends entirely on which cards are held. The figure this page publishes is what a simple heuristic achieves, and optimal play on the same paytable reaches about 99.5%. The gap between those numbers belongs to the player.
Format: Video poker · Jacks or Better · full-pay 9/6 · Demonstrates: A return that is a function of player skill, labelled by strategy
Play it freeRun the audit yourself
The machine
| Game | Five-card draw video poker, Jacks or Better |
|---|---|
| Paytable | Full-pay 9/6 — full house 9×, flush 6×, royal flush 800× |
| Strategy modelled | A simple heuristic, deliberately not optimal |
| Optimal play on this paytable | ≈ 99.5%, which this measurement does not reach |
| Basis of the figure | Measured over 60,000 hands at load, labelled by strategy |
The only game here where the player changes the number
Every other game on this platform has a return fixed by its constants. Video poker does not: hold the wrong cards and the same paytable pays materially less. That makes a single published RTP figure meaningless unless the strategy it assumes is stated alongside it, which is why the label here reads “under heuristic holds” rather than a bare percentage.
Full-pay 9/6 Jacks or Better returns about 99.5% to a player holding correctly. The heuristic in this build does not hold correctly. Both facts are on the page.
This page used to describe a different game entirely
Until this revision, /games/poker/ described Texas Hold’em — hole cards, a flop, heads-up play against the house, and a downloadable hand history the player owned. None of that was ever built. The game at this address has always been video poker.
The page has been rewritten rather than patched, and the discrepancy is logged in the project’s error ledger with the rest. A site whose argument is that it publishes only what it has built cannot keep a page describing a game that does not exist, and the length of time this one stood is part of the record.
The mathematics
Basis: measured. Sampled in your browser at page load and published with the interval that sample earned.
Full-pay 9/6 Jacks or Better — nine for the full house, six for the flush. The paytable below
is fixed and exact. The return is not, because it depends entirely on which cards are held,
and holding is the player’s decision rather than the machine’s.
| Hand | Pays |
|---|---|
| Royal flush | 800× |
| Straight flush | 50× |
| Four of a kind | 25× |
| Full house | 9× |
| Flush | 6× |
| Straight | 4× |
| Three of a kind | 3× |
| Two pair | 2× |
| Jacks or better | 1× |
Why this one is published as a range
The figure shown is what a simple heuristic achieves over 60,000
hands: 87–94%. Not a point estimate to two decimals — a range,
because a point estimate here would be false precision.
The royal flush pays 800× and lands about once in 40,000 hands. A 60,000-hand run that happens to
miss it reports a narrower interval than one that hits it — the sample looks more confident
exactly when it is least representative. So the error is floored at the royal’s own contribution,
pay × √(p(1−p)/N), which no run’s luck can move: 1.63 points against
the 0.66 the raw sample claimed.
Raising the sample instead was measured and rejected. Reaching ±1 point would need roughly 500,000
hands and about eleven seconds of the visitor’s time at load, which is a bad trade for a number that
is already honest as a range.
Optimal play on this paytable reaches about 99.5%. The heuristic here does not
come close, and the gap is yours to close — this is the only game on the platform where the player’s
own skill is a term in the equation, and the only one where the published figure is a floor rather
than a description.
What the gate does before the game loads
Nine states have enacted statutory bans on the sweepstakes casino model, and six more closed their markets by enforcement without passing anything. Every one of those instruments turns on redeemable value. This platform has none, so it stays lawful in all fifty — but the geo-gate still names the statute, because a compliance claim that cannot be checked against a citation is a marketing claim.
Select New Jersey and the gate blocks the lobby citing A5447/S4282. Select Oklahoma and it plays, with an amber banner recording that SB 1589 takes effect on 1 November 2026. The distinction between an enacted ban and a ban in force is the sort of thing operators discover late; here it is in the interface, fed by the same weekly-verified database as the state tracker.