An RTP figure without its rule set is unfalsifiable

Every live blackjack table in the market displays a return-to-player percentage, usually in the help panel, usually with a sentence explaining that the value is fixed and does not change. The wording is designed to sound like a measurement. It reads as though somebody weighed the game and wrote down the result.

For slots, that is roughly what happened, and there is no alternative. Reel weightings are proprietary, so the certificate is the only access anyone gets to the number. You either trust the lab or you do not play.

Blackjack is a different situation, and the industry has never quite reckoned with the difference. Blackjack's return is not a property of the game's name. It is a property of the rule set, and it can be computed from that rule set by anybody with a laptop and a weekend. Which means a published blackjack RTP is a falsifiable claim, and falsifiable claims should be checked.

A worked case

Take a live table advertising 99.44%. The rules, read off the in-game help panel on 11 August 2026, are specific enough to price exactly: eight decks, the dealer standing on all seventeens, a face-down hole card that is checked before the round proceeds, doubling permitted on any two initial cards, and naturals paid at 3 to 2.

The restrictions are where it gets specific. Splitting is capped at one per hand, a split hand cannot be doubled, and each split ace is dealt exactly one card and then stands. Surrender is absent from the document entirely.

Price that set from first principles, which means enumerating every opening deal, computing the dealer's finishing distribution behind each upcard, and taking the best of standing, hitting, doubling and splitting at every decision point. The answer comes out at 99.3943%, a house edge of 0.6057%.

That sits 0.046 of a percentage point below the advertised figure.

Why a rounding error is the wrong explanation

In money, 0.046 points is about 46 cents per $1,000 staked. Nobody should get excited about that, and this is not an accusation of anything.

It does matter methodologically, though, because it is larger than the model's own uncertainty. Running the same analysis with and without shoe-depletion effects moves the answer by roughly 0.002 points, so the gap is about twenty times the spread between reasonable modelling choices. Whatever it is, it is not noise.

The more interesting question is what would close it. Each blackjack rule has a known price, and the restrictions on this table are individually worth roughly:

Any single item on that list would more than cover a 0.046 gap. The help panel grants none of them.

So there are two readings. The certified number may have been computed against a slightly more generous variant than the one documented in the client, in which case the documentation is incomplete. Or the documentation is right and the advertised figure is optimistic by a few hundredths of a point.

Neither reading is fraud. Both are the same structural problem: the number and the rules are published in different places, at different levels of detail, and nobody is required to reconcile them.

Credit where it belongs

It would be dishonest to run this argument without noting what the same table gets right, because the rule that dominates all of the above is the payout on a natural.

Paying 6 to 5 instead of 3 to 2 costs the player about 1.36 points of edge. Paying even money costs about 2.26. Those figures are thirty to fifty times the gap under discussion, and 6:5 tables are now common enough that players have been trained to stop checking. A table paying 3 to 2 is handing back more value than this entire argument concerns, and the restrictions on splitting are a cheap price for it.

Precision cuts both ways. Any framework that flags a 0.046-point discrepancy also has to say clearly that a 3:2 natural is worth far more than the discrepancy, and that a player choosing on RTP alone should still take this table over most of its competitors.

The ask is small

The derivation behind these figures, along with the full rule matrix and the strategy chart the rules actually imply, is published at duel-5.com. That it had to be reconstructed by a third party from a help panel is the point.

Operators already hold the rule matrix. It exists in the game configuration and it was handed to the testing lab, because the lab could not have produced a certificate without it. Publishing it next to the percentage costs a table in a help file.

Two consequences follow immediately. Players get a number they can check, which is a stronger trust signal than any certificate logo. And an operator whose rules genuinely produce the advertised figure gets to prove it, rather than being lumped in with the ones whose figures came from a friendlier variant.

The current arrangement asks for trust in a claim that could simply be demonstrated. For a game whose return is fully computable, that is a choice, and it is the wrong one.