Castle Roulette (Duel): 100% RTP Wheel, Tested Live

United States: Duel does not accept players from the US. This page remains available as an independent game audit.

How to Play Castle Roulette

Despite the name, this is not European or American roulette. There are no numbered pockets, colours, dozens or outside bets, and the wheel has no 0 or 00. The name describes the spinning-wheel presentation; the bet is one or more multiplier tiers: 2x, 4x, 8x, 16x, 24x or 48x.

1. Choose a multiplier. Select the tier you want to back. You can back more than one in the same round.
2. Enter your stake. Bets are accepted during the countdown before the spin.
3. Wait for the result. One of the 48 wheel positions becomes the completed round outcome.
4. Collect the payout if your tier hits. The multiplier is applied to your stake and includes it: $10 on 2x returns $20 in total.
5. Open Verify if you want to audit the round. Duel exposes the inputs and JavaScript needed to reproduce the result.

What We Checked Live

We separate operator statements from things that can be counted, calculated or reproduced independently. Live checks: 24–25 Sep 2026. Documentation re-read: 23 Oct 2026.

Payouts from 2x to 48x

We counted 48 positions on the wheel: 24 for 2x, 12 for 4x, 6 for 8x, 3 for 16x, 2 for 24x and 1 for 48x. If every position is equally likely, each displayed multiplier is exactly the inverse of its probability, so every tier has the same theoretical expected return of $1.00 per $1 wagered while the fair paytable is active.

The multiplier includes the winning stake. A $10 bet paid at 2x returns $20 in total: the original $10 plus $10 profit. If that tier loses, the $10 is gone. Only the 2x payout was tested with a live wager; the other five rows are calculations from the same paytable.

Several bets in one round: subtract every stake from the winning payout. $10 on 2x plus $10 on 4x costs $20. A 2x result returns $20 and breaks even for the round; a 4x result returns $40 for $20 net profit; any other tier loses the full $20. Hitting one of your tiers is therefore not the same as finishing the round in profit.

Is It Really 100% RTP?

This was the key unresolved question in the earlier audit. Duel's general Zero Edge FAQ says that many Originals use a 0.1% base edge offset by 0.1% instant rakeback inside the allowance, while original slots are described as 100% RTP by design. Tiny wagers could not distinguish those models, because a 1.998x settlement and a 2.000x settlement can both display as $0.20 after cent rounding.

We therefore repeated the test at $10 on 2x. The result settled at exactly $20.00. A 99.9% base model would settle the 2x leg at $19.98 before any rebate. Because the observed base payout was the full face multiplier, the tested state does not use a 1.998x base settlement. Adding a separate 0.1% rebate on top of a true 2.000x payout would create a return above 100%, which is not the Zero Edge model Duel describes.

The wallet also displayed Rakeback available: $0 after the test. That is supporting context rather than the basis of the conclusion: a rewards field can be updated on a different schedule, stored separately or omitted from bet history entirely, so the absence of a visible reward line is weaker evidence than the settlement itself. The settlement already distinguishes 2.000x from 1.998x at cent precision.

For platform-level context, see the Duel Casino audit. The broader allowance rules are covered in zero-edge allowance explained.

What 100% RTP Does Not Mean

100% theoretical RTP does not make a short session risk-free. It describes long-run pricing, while individual results can still vary sharply; see can you lose with 100% RTP? for the separate variance discussion.

How to Audit the Table

1. Count the wheel positions and confirm the six-tier distribution.
2. Compare each multiplier with the inverse of its observed probability.
3. Use a wager large enough to distinguish exact face pricing from a 0.1% haircut after normal display rounding.
4. Check the Zero Edge allowance and post-limit wording.
5. Open a completed round and reproduce the result from the published verification code.

Compared with Roulette

Round #986542 Reproduced

Duel's verifier publishes both the round inputs and the JavaScript used to calculate the wheel position. For completed round #986542, the Results tab itself mapped Roll 39 → 2x; the independent calculation below reproduces position 39.

Reproducing Roll 39

The server seed is decoded from hex into bytes: those bytes are both the SHA-256 commitment input and the HMAC key. For the message, the verifier decodes the drand signature bytes as UTF-8 (invalid sequences are replaced by the decoder), appends :0, and encodes that text as UTF-8. It then computes HMAC-SHA256, takes the first eight hex characters and applies modulo 48.

keyBytes   = hexToBytes(serverSeed)
commitment = SHA256(keyBytes)
drandText  = hexToUtf8String(drandSignature)
message    = utf8Encode(drandText + ":0")
hmac       = HMAC_SHA256(keyBytes, message)
value      = parseInt(hmac.slice(0, 8), 16)
roll       = value % 48

Implementation notes: the hexToUtf8String step must be reproduced literally — signing the raw drand bytes or the hex string itself produces a different roll. The HMAC key must be the seed bytes: using the 64-character hex text as the key gives Roll 27 and a commitment that does not match.

What This Verification Proves — and What It Does Not

The reproduced round establishes two concrete facts. First, the revealed server seed is consistent with the published commitment: hashing the revealed seed gives the value shown in the Verify panel. Second, when the published calculation is run with the disclosed server seed and drand input, it returns the same wheel position shown by Duel.

Matching the verifier input to public Quicknet round 32492923 establishes which published beacon value was used. That match alone does not establish when the server-seed commitment was published or when the beacon round was selected. We did not capture the commitment with an independent timestamp before the relevant drand value became public; a stricter audit would record the commitment before the spin and then verify that the later public randomness was not yet available at that moment.

The Live Show

Unlike every other Duel Original, this one is not software-only. The round is streamed from Duel's studio: a presenter spins a physical wheel on camera, the betting panel with the six tiers sits on the same screen, and the Results and Verify tabs for every completed round are one tap away. Bets close when the countdown ends, the wheel is spun, and the winning segment is settled immediately.

The settings panel lets you replace the stream with a virtual-wheel animation while keeping the same betting interface and completed-round data. This is a presentation choice: the provably fair calculation is handled separately through the round verifier, so the virtual view is as auditable as the stream.

Reading the Last 100

The game shows recent outcomes and a Last 100 count for all six tiers. The counter is useful for picking rounds to verify and for seeing how uneven a fair wheel looks over a short run. It is not a prediction tool. Over 100 independent spins the counts scatter widely around the long-run proportions, and a run that looks "off" is usually inside the normal range:

Ranges are binomial: ±2 standard deviations around the expected count, so roughly 19 out of 20 fair runs of 100 spins land inside them. A Last 100 with no 48x at all, or with 60 hits on 2x, is ordinary. Only a count far outside these ranges, repeated across many independent windows, would justify looking harder at the wheel — and the place to look is the round verifier, not the streak. Third-party trackers that log droughts and "overdue" tiers describe the past; they do not change the probability of the next independently generated round.

Which Multiplier to Pick

If the fair paytable is active, no tier has a higher expected value than another. What changes is variance.

Betting several tiers changes the distribution of session wins and losses but does not create a mathematical advantage, and round history does not make a multiplier "due". The game suits players who want a handful of simple choices instead of a roulette betting grid, players who care about verifiable rounds, and — on the 16x, 24x and 48x tiers — players who accept long dry spells in exchange for larger individual payouts.

Sources and Methodology

Operator-specific claims are separated from our own observations and calculations. Primary references:

• Duel Zero Edge FAQ — general allowance and post-limit framework.
• Duel Provably Fair — general fairness documentation; the game-specific verifier was used for the reproduced round.
• drand HTTP API documentation — public randomness beacon background.
• UK Gambling Commission: Return to Player — theoretical RTP as a long-run metric.

The wheel count, EV calculations, $10 settlement test and round reproduction are first-party checks by 100RTP.games. The payout conclusion applies while the Zero Edge allowance was active; the post-limit pricing rule is documented by the interface but was not measured with real wagers. The reproduced round demonstrates that the disclosed verifier works for that completed event, not that every future implementation change is automatically covered by the same code.

Evidence types are kept separate in the wording. "Operator-stated" is information shown by Duel that we did not reproduce independently, such as the exact post-limit scaling formula. "Formula-level verified" means the result follows directly from the counted wheel and displayed multipliers. "Independently verified" is reserved for checks we reproduced ourselves: the 48-position count, the $10 settlement and the completed-round calculation.

Related Audits

• Duel Casino audit
• 100% RTP casino games audit
• Zero-edge allowance explained
• How to verify provably fair results
• How zero-edge mechanics work
• Duel Dice audit
• Duel Crash audit

Frequently Asked Questions

Bottom Line

Inside the Zero Edge allowance, Castle Roulette paid the full face multiplier: the 48-position wheel matches the 2x–48x paytable, the live 2x payout settled at exactly face value, and round #986542 reproduces from Duel's published seeds. Two things stay operator-stated: the exact house edge after the $50,000 daily allowance, and the timing of the server-seed commitment, which we did not timestamp independently.