Rakeback Test Methodology: How We Verify Duel’s 50% Instant Rakeback

Protocol version 2.2 · Last updated 17 July 2026 · Applies to all pages in the Live Rakeback Tests series

Every slot test we publish follows the protocol on this page. It exists so that any reader can check our work: what we claim to verify, exactly which numbers we record, how we decide that a result passed or failed, and where the method's limits are. Each test page states the protocol version it ran under; the changelog at the bottom explains what changed between versions.

What these tests verify — and what they cannot

A hundred spins cannot produce a meaningful estimate of a slot's long-run RTP. On high-volatility games the spread of 100-spin outcomes is enormous — sessions in our results table have returned anywhere from 26.5% to 593.2% of stakes — so any "measured RTP" from a short session is noise presented as data, and we never publish one. What a short session can verify is a deterministic claim: that a fixed share of a game's house edge is credited back as turnover settles, win or lose. If that is true, the accrual is a predictable function of turnover, and a modest test either lands on the prediction or it doesn't. That is the claim under test, and the only one.

Terminology we hold ourselves to

Badge reference RTP

The figure Duel's rakeback badge shows as "Original RTP". The advertised rebate is built from this number — and, as our tests found, it is not always the RTP of the mode you actually play.

Active-mode RTP

The published RTP of the configuration actually used in the test — base game, ante, feature buy or super spins — taken from the in-game rules pages where we play, never from a provider's press materials.

Badge rakeback rate

(1 − Badge reference RTP) × advertised rebate share. The flat per-game rate our tests check the counter against.

Active-mode net edge

(1 − Active-mode RTP) − verified badge rakeback rate: the residual theoretical edge after the rebate, conditional on the badge remaining active and the accrued amount being claimable. When the badge quotes a richer mode than the one you play, this is higher than half your mode's edge.

Conditional effective return

Active-mode RTP + verified badge rakeback rate. We use this term instead of calling the combined figure an RTP, because rakeback is not part of the game engine: it accrues outside the round and requires claiming. Duel's interface labels its own aggregate "Duel RTP"; we quote that label only as what the interface says.

Session return

What a block actually paid back divided by what it wagered — a descriptive result of one block, reported for transparency, never an estimate of anything.

What we record

Each test is one block of 100 settled paid wagers at a fixed stake in a declared mode — no stake changes, no parallel wagers, claims, deposits or withdrawals during the block. Free spins triggered inside a bonus belong to the wager that bought them and add nothing to turnover. Before the block: all prior wagers settled, the rakeback counter claimed to zero (or its baseline separately recorded), the page refreshed and the zero re-checked. Fixtures captured: the badge tooltip, the rules pages with every published mode RTP and the bet limits, and the pre-block counter.

The core record is five figures: starting balance, settled turnover, total winnings, ending balance, and accrued rakeback. Turnover comes from the batch panel's TOTAL BET (Pragmatic's Hyperplay, Nolimit's Action Spins), cross-checked against wagers × stake; a mismatch halts the test until explained. The batch panel also separately logs every spin — same interface as the balance, so a consistency check rather than an independent source, and archived as evidence. One identity must hold before anything else is evaluated:

Where the expanded rakeback counter is available, its full displayed precision is the primary accrual measurement; the compact cent display serves as a secondary control. The cent-only protocol applies only where the expanded reading is unavailable or impractical, and the page then says so in its limitations. Where tests run back to back, we also reconcile the chain between them: each test's starting balance should equal the previous test's ending balance plus its claimed rakeback.

How a result is graded

The compact counter displays whole cents, so grading works on display bins, not on a distance tolerance. From each candidate model (the badge reference rate, and where relevant the active-mode rate) we compute the range of accruals consistent with the published RTP's own rounding — a figure displayed as 96.50% at two decimals stands for roughly 96.495–96.505% — and convert that whole range into the set of cent values it can display as. A result is Matched at displayed precision when the observed display falls inside the model's display set, and the ledger check holds. When the expanded counter was captured, the stricter test applies instead: the observed rate (accrued ÷ turnover) must fall inside the interval implied by the published RTP's precision — Matched within published RTP precision. A result earns a Replicated flag when an independent second block reproduces the rate. It is Inconclusive when two or more candidate models produce the same display set on the tested turnover — we say so rather than picking a winner. It is Not confirmed when the observation falls outside every candidate's admissible range, or the ledger fails to reconcile; a Not confirmed result triggers one independent repeat block before publication, and only a Reproduced discrepancy is reported as a finding.

Two candidate figures can be told apart whenever their display sets differ on the tested turnover — which can happen at a difference of a single cent, or less, when the predictions straddle a display boundary. That is exactly how our Sweet Bonanza 1000 mode audit rejected the active-mode model. One honest caveat the series itself surfaced: in our directly cross-checked readings the compact counter truncated to cents, but one reading was ambiguous about the convention, so where a conclusion would hinge on truncation versus rounding of the compact display alone, we grade Inconclusive — the expanded reading settles the question whenever it is captured.

Test types

A standard base test is one block in a declared mode, graded against the badge reference rate and, where the figures differ, the active-mode rate. A replication audit adds a second independent block at a different stake or with a re-zeroed baseline; it tests reproducibility of the rate, not mode behavior — our Gates of Olympus 1000 pilot is one, and note that its base and ante RTPs are identical, so it cannot distinguish accrual models. A mode audit runs separate blocks in modes whose published RTPs genuinely differ, with predictions that land in different display bins — the only design that can decide between a flat badge rate and active-mode accrual. Our Sweet Bonanza 1000 test is the series' mode audit.

The observed best-mode pattern

Two distinct observations run through the series, with different evidential weight, and we keep them separate. Structural: across all 11 inspected games, the badge reference RTP matched the highest RTP among the modes published on the game's own rules pages — checkable from screenshots alone, 11 of 11. Financial: in the subset where candidate figures separated at display level on our turnover — one full mode audit (Sweet Bonanza 1000) and three single-mode tests with resolvable badge-versus-rules gaps (Le Bandit, Wanted Dead or a Wild, Zeus vs Hades) — the displayed accrual came out consistent with the badge reference rate rather than the active-mode figure: 4 of 4 resolvable checks. The remaining games either have agreeing figures or gaps too narrow for their display sets to differ, which is consistent with the pattern but does not test it. It remains an observed pattern, not a documented Duel policy: every new test doubles as a check on it, and we will report the first counterexample as prominently as the confirmations. The player-facing consequence: when you play a mode below the game's best, your active-mode net edge exceeds half your mode's edge — hundredths of a point that compound with volume. The full picture is in the series results table.

Limitations

All results come from one account, on stated dates, in USDT, with the 50% Instant Rakeback badge active. Rates could differ by account, region or time, and Duel's terms can change after publication — every test page carries its date for exactly that reason. Counter and balance readings come from the site's own interface; they reconcile with each other to the cent, but we do not intercept API traffic to audit the backend independently. Conclusions cover the accrual formula, plus one operational check performed once for the series: claimed rakeback was subsequently withdrawn from the account in USDT, with no wagering requirement applied. Nothing on these pages predicts session outcomes, and nothing here is a reason to gamble.

Funding and conflicts of interest

These are self-funded, documented and reproducible tests. Every wager was paid for by the site owner; Duel provided no balance, fee or other compensation for any test in the series. Pages may contain affiliate links, and affiliate relationships do not alter the protocol, the formulas or a single published number — the archived screenshots exist so you don't have to take that on trust.

Changelog

The complete results table for the series lives on our rakeback and house edge hub, and each game's page carries its full data and evidence. Questions about the protocol are welcome — a methodology that can't survive questions isn't one.