RNG Pillar·Chapter 6

Live dealer vs RNG

In an RNG title the outcome is generated in software; in a live-dealer game it comes from the physical world — a real shuffle, a real wheel, a real ball — and a camera reads it into the platform. That single difference moves what a laboratory certifies: from a software generator to physical equipment, dealer procedure and the accuracy of the recognition that digitises the result.

By Vadym Zelinskyiverified 2026-09-18Current · 100%

Two games can feel almost identical — the same roulette wheel, the same blackjack table — and yet answer the question “where did that outcome come from?” in completely different ways. In an RNG title the result is generated inside software. In a live-dealer game it comes from the physical world: a real dealer shuffling and dealing real cards, a real wheel, a real ball. This chapter is about that one difference, and what follows from it — because it quietly relocates the thing a laboratory has to certify.

Two places randomness can live

The RNG side is the subject of its own chapter, so here it is only the contrast that matters: in a software title the outcome is a number from a generator — a pseudorandom algorithm, usually a cryptographically secure one, seeded from a hardware entropy source — mapped to a symbol without bias. All the unpredictability lives in the code and its seed.

A live-dealer game moves the source out of the computer entirely. GLI-19, the standard for interactive gaming systems, frames its live-game rules around wagers placed on games “conducted by a gaming attendant (e.g., dealer, croupier, etc.) and/or other gaming equipment (e.g., automated roulette wheel, ball blower, gaming device, etc.) in a live game environment”, where “the entire process is viewed by all players through real-time remote audio and video feed”. The randomness is a physical event you can watch happen. Nothing is being generated; something is being observed.

How a physical result becomes a digital one

A camera sees a card land face-up or a ball settle in a pocket; the platform needs to know that it was the seven of clubs, or pocket 17, to settle every wager automatically. Bridging the two is a recognition problem — the studio equivalent of optical character recognition, reading the rank and suit of a card or the number of a wheel pocket into the game state (many studios add RFID tags in cards and wheels as a second, tamper-resistant read, a set-up covered in the live-dealer studio tech guide). GLI-19 calls this layer “automated recognition” and puts a hard requirement on it: the software shall “ensure a very high degree of accuracy in identifying and reporting the game outcome data to the Gaming Platform”.

Notice how this changes the integrity question. For an RNG title you ask whether the generator is sound. For a live game you ask whether the physical equipment is fair and whether the recognition read it correctly — and the standard treats the physical event, not the camera, as the source of truth. The digitised outcome “shall match the outcome generated by the physical randomness device”, there must be a manual mode to correct a misread — “where the device misreads a card, the position of the ball, etc.” — and where the recognised data and the physical result disagree, “the physical randomness device’s outcome shall be considered correct”. The camera reports the wheel; it never overrules it.

What gets certified — and it isn’t a software RNG

For an RNG title the certification target is the generator and the game maths: the RNG requirements of GLI-11 and the statistical evaluation that a certificate actually asserts. For a live game there is no software RNG to certify — so the target becomes the physical chain instead. That is why the standards library carries separate documents for it. GLI-29 covers the physical card-handling equipment directly, defining a card shuffler as “a device that is designed, at a minimum, to have the capability to randomly rearrange a deck or decks of playing cards to eradicate any patterns”, and GLI-19’s live-game section governs the studio process, the dealer procedure and the recognition software.

The honest way to say it: an RNG title is certified as software; a live game is certified as an apparatus and a procedure. Same laboratories, same 99% statistical bar on the outcomes, different thing on the bench. GLI-29 makes the split explicit for a shuffler, which pairs mechanism and code — “the RNG and the physical mechanics of the shuffling device will mutually result in the production of random card outcomes” — and then requires those outcomes to be, in the same words the RNG standards use, “statistically independent… and… unpredictable”.

Different fairness profiles, same goal

It is tempting to rank the two — to decide a physical wheel must be “more real” than a line of code, or that software must be “cleaner” than a piece of felt. Neither holds. The two approaches simply move the trust to a different place: an RNG title asks you to trust a tested generator; a live game asks you to trust tested equipment plus an accurate optical capture. Each has a matching failure surface, and each is audited precisely because it does. A live deck or wheel is only as fair as the hardware and the procedure around it — which is the whole reason GLI-29 exists and why a shuffler must be shown to produce independent, unpredictable results rather than assumed to. The destination is identical to the RNG pillar’s: outcomes that cannot be steered against a player in normal certified play. Only the road there differs.

Where the two blend

The line is not always clean, and it helps to name the hybrids so they are not mistaken for something they are not. Some studio games look live but are pure RNG — the “First Person” tables, whose 3D presentation borrows the studio aesthetic while the outcome comes from a certified generator, not a physical wheel (the First Person & hybrid tables guide draws that boundary in detail). Others run the other way: a genuinely live host, dealing a real game, with an RNG-drawn element bolted on — the multiplier spots in Lightning-style tables are selected by a random generator on top of the physical spin. The rule of thumb is simple: wherever an outcome is drawn in software, that portion is certified like any RNG — against GLI-11 and the statistical batteries — and wherever it comes off a table, it is certified as equipment and procedure. A hybrid is just both jobs done at once.

Common questions

Is a live-dealer game more fair than an RNG slot?

Neither is inherently more fair — they place your trust in different things. An RNG title generates the outcome in certified software; a live game takes the outcome from the physical world, and GLI-19 recognises exactly that, allowing live games to "utilize physical randomness devices … to generate game outcomes." A live wheel or shuffled deck is only as fair as the equipment and procedures behind it, which is why the physical apparatus is itself certified (for example under GLI-29 for card shufflers and dealer shoes) rather than taken on faith. Both routes are audited to the same goal: outcomes that are independent, correctly distributed and not manipulable in normal certified play.

If a real dealer deals real cards, what is there to certify?

The target shifts from a software generator to the physical chain. Laboratories examine the equipment (GLI-29 defines a card shuffler as a device able to "randomly rearrange a deck or decks of playing cards to eradicate any patterns"), the dealer and studio procedures, and the software that reads the physical result into the platform. GLI-19 requires that automated-recognition software "ensure a very high degree of accuracy in identifying and reporting the game outcome data." So the integrity question moves from "is the generator sound?" to "is the equipment fair, the procedure controlled, and the recognition accurate?"

What happens if the camera misreads a card or the ball?

The standard plans for it. GLI-19 requires a manual operation mode "to allow for corrections of an erroneous result (where the device misreads a card, the position of the ball, etc.)," and makes the physical result authoritative: "Where a discrepancy between the physical randomness device and the game outcome data exists, the physical randomness device’s outcome shall be considered correct." The camera and recognition layer report the outcome; they do not get to overrule what physically happened on the table.

Sources (4)

Education, not advice. This chapter explains how the random number generators behind casino games are built, tested and certified so you can judge the claims made about them honestly. It is not a system, and nothing here treats gambling as a way to make money — over enough play the mathematics favours the house. 18+.

Next in the pathRNG in crash games