Nobody had published what this wheel is actually made of. The game reports a raw position for every spin and colours every past result by what it paid, so we lined the two up and recovered the whole layout: 48 segments, six tiers, and a price for each.
The wheel as the game indexes it. Position 0 is the lone 48x; the blocks widen as the multipliers fall, ending in 24 segments of 2x.
Castle Roulette gives you six buttons, 2x through 48x, and a wheel on a video feed. Because the wheel is streamed video rather than something the browser draws, you cannot count its segments in the page. Both write-ups that cover the game hedge on exactly this point: the segment split is quoted from promotional material, and whether the live wheel matches it is left open.
It turns out the answer is sitting in the game itself, in two halves. The first is a numeric wheel position the game records for every finished spin. The second is the strip of past results along the top of the table, where each spin is drawn in the colour of the tier it paid, and the operator names its own tally elements count-two through count-forty_eight, which is what pins each colour to a multiplier without guesswork.
Line the position up against the colour, round by round, and the layout stops being a claim. Positions run 0 to 47, and they are grouped in six unbroken blocks rather than scattered around the rim. Across the 28 spins where we held both a position and a colour, the layout below predicted the tier 28 times out of 28.
The whole wheel, by position. Each tier pays exactly the inverse of its own chance, which is why the last column reads the same all the way down.
| Pays | Positions | Segments | Chance | As a % | Odds | Fair payout | Returns |
|---|---|---|---|---|---|---|---|
| 2x | 24–47 | 24 | 1/2 | 50.00% | 1 in 2 | 2x | 100.0000% |
| 4x | 12–23 | 12 | 1/4 | 25.00% | 1 in 4 | 4x | 100.0000% |
| 8x | 6–11 | 6 | 1/8 | 12.50% | 1 in 8 | 8x | 100.0000% |
| 16x | 3–5 | 3 | 1/16 | 6.25% | 1 in 16 | 16x | 100.0000% |
| 24x | 1–2 | 2 | 1/24 | 4.17% | 1 in 24 | 24x | 100.0000% |
| 48x | 0 | 1 | 1/48 | 2.08% | 1 in 48 | 48x | 100.0000% |
| Whole wheel | 0–47 | 48 | 1 | 100% | n/a | n/a | 100.0000% |
Half the wheel pays 2x, a quarter pays 4x, an eighth pays 8x, and so on down to a single segment paying 48x. Chance times payout comes to 100.0000% on all six tiers. There is no rounding fudge and no shaved multiplier anywhere on the rim.
We measured all 27 Plinko payout tables the same week and not one reached break-even; they keep between 0.5% and 1.1% of stake. Same casino, same Zero Edge badge, and the maths lands in a completely different place.
The game ships a settings block naming a house edge floor of 0.1% and a ceiling of 5%. The segments we mapped price at zero. Both things are true at once, and the section below works through why that is not a contradiction.
Above the table the game keeps a running count of how often each tier has landed in the last 100 spins. Decoding 100 raw positions through the layout should reproduce it.
| Source | 2x | 4x | 8x | 16x | 24x | 48x | Total |
|---|---|---|---|---|---|---|---|
| Our decode of 100 positions | 50 | 24 | 13 | 7 | 4 | 2 | 100 |
| The game's own Last 100 | 50 | 25 | 13 | 6 | 4 | 2 | 100 |
| Difference | +0 | -1 | +0 | +1 | +0 | +0 | +0 |
The two rows agree to within 2 across six tiers, and the residual is the one or two spins that landed between reading the history and reading the counter. Note which tier this settles: 24x sits at positions 1–2, and those never came up during the window where we could also see a colour. The counter is what confirms them, holding steady at 4 while our decode of the same window also returns 4. Of the 48 positions on the wheel, 45 showed up at least once in 100 spins.
Alongside the spin history the game serves its own settings block. Most of it is plumbing: which server carries the video, how long payouts are held back, the smallest tip a host will take. Two fields are not plumbing at all. One is named min_house_edge_limit and holds 0.1. The other is max_house_edge_limit and holds 5.
Read as percentages, those are returns of 99.9% at one end and 95% at the other. On $10,000 put through the wheel that is a spread from $10.00 to $500.00. The floor is the more telling number, because 99.9% is precisely the figure one of the two existing write-ups reached for when it hedged its description as a base game plus a small return layer. That hedge looks well judged: the number it guessed at is sitting in the config.
None of which makes the segments we mapped wrong. Those are the multipliers the wheel is paying, and they price at zero edge. What the config shows is that the wheel is not the only thing in play, and that the operator has given itself a band to work in rather than nailing the game to zero. Alongside the fields sit an exposure limit of 5% and a bankroll figure of 50,000,000, which is the language of risk management, not of marketing.
So the honest reading is narrower than either the promotional line or a debunking. The paytable on this wheel really is fair, unusually so for a live game show, and better value than any roulette table you will find. It is also configured with room to be up to 5% worse than that, and Duel does not publish which end of its own band any given spin sits at.
This is the part of Castle Roulette that deserves more credit than it gets, and it is not mentioned anywhere else we could find. Every spin records the round number of a drand beacon, the public randomness service run by the League of Entropy. drand publishes a fresh unpredictable value on a fixed schedule, signed by a threshold of independent operators, and no single participant can steer or withhold it. Duel is not rolling its own dice here. It is quoting a number the whole internet can look up.
Two things fall out of that, and both check out. Consecutive spins sit exactly 10 beacon rounds apart, and the quicknet chain publishes every 3 seconds, so the wheel is pinned to a 30-second heartbeat. That is the 30-second cadence we timed independently, arrived at from the other direction. Then we took 8 of the beacon rounds named in the history and pulled them from the beacon's own public API: all 8 matched the value Duel had published, byte for byte.
The commitment side holds up too. Before each spin the game shows a hash and afterwards reveals the seed behind it. Hash the revealed seed and you should get the hash you were shown, which is what proves nothing was substituted once bets were down. That reproduced on all 100 of the 100 rounds we checked, once we worked out that the hash covers the seed's raw bytes rather than the text you see.
Now the gap. Having the beacon value and the revealed seed should let you recompute the wheel position yourself, and it does not, because the function that combines them is not documented. We tried 27 standard constructions and none of them reproduced the positions. The fairness panel that might spell this out sits behind a login, so an unregistered visitor cannot inspect it at all. The result is a real but partial guarantee: you can prove the inputs were fixed in advance and that one of them came from outside the casino, and you cannot prove they produce the number the wheel showed you.
Our verification walkthrough covers the commit-and-reveal scheme as it applies across the Originals, where the algorithmic games such as Dice and Crash are considerably easier to check end to end.
With every tier priced at 100%, no button is better value than another. What changes is how often you are paid and how hard the swings are.
| Tier | Lands | Average wait | In minutes | Median dry spell | No hit in an hour | Swing per $1 |
|---|---|---|---|---|---|---|
| 2x | 50.00% | 2 spins | 1 min | 1 spins | 0.0% | 1.00 |
| 4x | 25.00% | 4 spins | 2 min | 2 spins | 0.0% | 1.73 |
| 8x | 12.50% | 8 spins | 4 min | 5 spins | 0.0% | 2.65 |
| 16x | 6.25% | 16 spins | 8 min | 11 spins | 0.0% | 3.87 |
| 24x | 4.17% | 24 spins | 12 min | 16 spins | 0.6% | 4.80 |
| 48x | 2.08% | 48 spins | 24 min | 33 spins | 8.0% | 6.86 |
The last column is the useful one. On a fair bet paying m, the swing per dollar staked works out to the square root of m minus one, so it climbs from 1.00 on the 2x to 6.86 on the 48x. Backing the crown is 6.9 times as volatile as backing the steady tier while returning exactly the same amount over time. Spreading a stake across several tiers smooths the ride and changes nothing about the return, because every part of the spread is priced identically.
The word describes the spinning, not the maths. Set beside actual roulette the differences run all the way down.
| Castle Roulette | European roulette | American roulette | Crazy Time | |
|---|---|---|---|---|
| Wheel positions | 48, indexed 0 to 47 | 37 pockets | 38 pockets | 54 segments |
| What you back | One of six multiplier tiers | Numbers, colours, ranges | Numbers, colours, ranges | Numbers plus bonus rounds |
| Zero pocket | None | One | Two | None |
| Best bet returns | 100.00% on every tier | 97.30% | 94.74% | Around 96% |
| House edge on the layout | None | 2.70% | 5.26% | About 4% |
| Randomness | drand public beacon | Physical wheel | Physical wheel | Studio wheel |
| Can you check a round | Commitment yes, position no | No | No | No |
Roulette keeps its edge with pockets that pay less than their odds deserve: a single number on a European wheel lands once in 37 and pays 35 to 1, and the American wheel adds a second zero to widen the gap. This wheel has no equivalent. There is no zero segment, and each tier pays its exact inverse. If you arrive expecting roulette you will find the bet types missing; if you arrive expecting a game show money wheel you will find one priced far better than the genre norm.
The clock is the thing most likely to cost you here, given the paytable will not. A spin lands every 30 seconds and the window to place a bet is 13.6 seconds of that, under half the round. That is 120 decisions an hour if you play every one, on a game where no decision improves your return.
18+ · Gamble responsibly · Wheel measured at duel.com/castle-roulette on 8 August 2026 · Affiliate link disclosed