How We Review Slots

The short version

We test in demo mode

Every figure we publish comes from a free-play demo session run on the provider's own client. We don't post real-money results or "I deposited X and withdrew Y" stories, because those prove nothing about the game and everything about luck on the day.

Provider demos are built to run the same game rules and mechanics as the real-money version. Operators can still deploy different RTP configurations, and we cannot independently verify that every demo and live build is identical - which is one more reason we publish observed numbers instead of promises. The one number a demo can't fix for you is RTP, because the live RTP is set by each casino operator - many games ship in several RTP versions and the casino chooses which one to run. Always check it in the game's info screen before you bet real money.

Our process, step by step

1. Read the game first. The in-game Help, rules and paytable screens are our primary source for mechanics, jackpot values and bet limits. We read every page before forming an opinion.
2. Run a controlled session. At least 1,000 spins, and recent sessions run 1,600 to 3,000, at a fixed stake set before the first spin, on the official provider demo where possible, with the balance logged on every spin. Each session report publishes a checkpoint table read from that log at scheduled intervals plus every feature trigger, the high and the low; the full ledger and the frame archive are retained. Any session that departs from the fixed stake is labelled as such in its report.
3. Count by money, pool with care. Observed return is total returns divided by total stakes, so a mixed-stake session is measured by money and says so. Sessions are pooled only when every one of them provides complete wager and return totals; otherwise the review shows the portal session alone. Purchased features are counted separately from natural triggers.
4. Log the session data. Spin count, balance checkpoints, base-game hit frequency, longest no-win run, and every bonus trigger with what it paid.
5. Cross-check every spec. RTP, volatility and max win are verified against the provider page and independent databases, and we tell you when sources disagree and which value we used.
6. Finish the session before publishing. No review goes live on a partial session, and no number is invented. If the data is too thin for a claim, we say so.

Where our numbers come from

When sources conflict, the one higher in this list wins, and we note the conflict in the review:

1. The build we tested - its rules and info screens: the RTP where the build prints one, all symbol rules, payout values and the stated maximum. A studio can ship one game at several return settings, so for the session we logged the figure inside the build outranks the provider's page.
2. Official provider game page or game sheet - release date, volatility, the return settings a game ships at, and the RTP when the build prints none.
3. SlotCatalog / AskGamblers - when the provider doesn't publish a figure.
4. LCB / Slotozilla - secondary cross-check when sources disagree.
5. Session observation - behavioural notes that aren't in any spec sheet.

Our own session figures come from logged demo sessions with balance checkpoints, and every one of them has a published session report with the checkpoint table, the bonus register and the closing arithmetic, for example session report LAVB-20260730-01. The spin-by-spin logs are available on request, and one of them is published as a CSV file for anyone to check: session report RBGD-20260811-01.

How we score

Every review carries a score out of 10. It sits in the Quick verdict at the top of the page, and the review's closing section names what earned marks and what lost them. It is an editorial score of a game's mechanism and of how openly the build states its terms, not a measurement of the game's maths and not advice to play. The score is a judgement and not a formula: we do not award points per reason and add them up, but we can say which reasons decide it. The first five below come up in every score published so far, and the sixth is a matter of taste that we add openly.

What moves the score

1. The mechanism. Whether the central idea is real rather than a decal, and whether its clever piece sits where the player spends the session. This is the reason that carries most.
2. Disclosure. Which figures the build in front of the player prints: its return, its feature rate, the odds or weights behind its symbols. A game that shows its RTP on a rules page, from a studio that publishes its configurations, earns credit. A studio that certifies a return and then leaves it out of the build loses marks, and this is the most frequent deduction on the site. It is charged to the build as shipped, not to the studio's reputation. No build we have tested so far prints the weights behind its symbols. On its own that gap has cost no game a mark, and a closing section names it only where it completes a build that prints nothing else, as Fortune Rabbit's does.
3. Legibility. Whether a player can follow on screen why a spin paid what it paid: a trigger that can be counted, multipliers visible before a win resolves, a price list printed in full. This reason is about the process, and disclosure is about the figures.
4. The headline and what backs it. How the top end is sold, set against what the rules say about reaching it. The charge here is for a promise without a price. A jackpot or a named feature that the splash screen, the title or a panel above the reels sells as something to be won loses marks when the rules give it no route, attach a condition the pitch leaves out, or print no odds for it. A maximum that the rules print as a cap, the point where a round is cut off, is a limit on the player and not a promise, and it draws no charge however large it is. A feature or a ladder rung that the game's own printed odds put at one spin in millions is marked as decoration, and the review gives those odds. What our pools did or did not meet is reported beside the headline with its sample size, as an observation and never as the ground for a charge.
5. The base game between features. How often a spin pays anything, how often it pays more than the stake, and how the feature clock runs, read as the range its count supports. We prefer a base game that takes part in the session over one where the spins between features only collect a toll, and we mark accordingly. That is a preference and we state it as one: a player who wants rare, large events may weigh the same figures the other way.
6. Presentation. How the game looks, moves and sounds, and whether the theme is carried by the mechanism or only painted over it. This is taste, and we label it as taste. Reviews are written by people, and a machine that is a pleasure to sit with for two thousand spins earns something for it, as one that grates loses something. It works as a tie-breaker: it settles the order between games that the first five reasons leave close, and it never lifts a game past a neighbour whose mechanism or disclosure is plainly stronger. The closing section says in a sentence what we liked or disliked, so a reader who disagrees can set that sentence aside. Presentation enters with the reviews published after 17 September 2026 and with any re-test of an older game.

One absence is charged once. When a missing figure could be filed under two reasons, the closing section names it under one of them and the other stays silent on it. Fortune Rabbit's unprinted tile weights are also the odds of the 5,000x its loading screen sells, and they are counted against it once, under disclosure.

What the sample can and cannot carry

A session runs at least 1,000 spins at a fixed stake, recent ones 1,600 to 3,000, and the figures it produces are not equally firm. The share of paid spins that return anything is the firmest. One stake is one observation, and free spins are not counted as spins. On 2,000 such spins the share carries a margin of about two percentage points either way at the usual 95% level, and in the four pools where all three logs print that share it stayed within three points from one log to the next. Shares of that kind inform the score, read with the length of the log beside them. The longest run without a payment is a single extreme and moves far more, 45, 48 and 63 spins across three logs of one game, so it illustrates a frequency and adds nothing to it.

A count of rare events gives a range and not a rate. Eleven free spins rounds in 5,425 spins place a trigger between one spin in 275 and one in 1,000 at the same level. Every figure in that range is a slow clock, so the review may call the clock slow, and the pooled one in 493 is printed as arithmetic on eleven events and not as the game's rate. A tier that stayed shut through 48 rounds could open once in twenty rounds or once in a million, and the log cannot tell which, so its absence is reported and never charged. Both margins assume that one spin does not depend on the last, and where a meter or a bank carries state across spins they are wider than stated. Where a studio prints a feature rate we set the log's rate beside it and do not replace it. Thin evidence lowers our confidence in a statement, and the review says so; it does not lower the game's score.

What does not move it

• The session return. Lava Balls holds the top score, 8.1, with a pool that closed at 88.85% and two sessions out of three in the red. Speed Winner sits at 7.2 with a 149.43% session in its headline. The return is printed beside the score and never folded into it. How far a session can land from the stated figure depends on its length and on the game's volatility, which is the subject of observed vs stated RTP.
• The size of the stated return. Stated returns on our shelf run from 95.62% to 96.75%, and the games at both ends are scored on the same reasons. Studios ship one game at several return settings and the casino picks one, so the figure a reader meets may not be the one we tested, and where a build prints no return the figure comes from sources we cannot confirm. What is scored is whether the build prints its return, which is the second reason above. The figure itself is printed in every review, with its source.
• The size of the maximum win. A 100,000x cap earns nothing on its own. What counts is whether the build tells the player how that figure can be reached and how often, and that question belongs to the fourth reason above. The one game marked down near this point is Burning Wins, whose best line pays 60x, and the mark is for a base game with long droughts and little above them, which is the fifth reason.
• The theme as a label. A genre, a mascot or a licence earns nothing by itself. How the theme is carried belongs to the sixth reason above.

The scale and its tenths

Scores published so far run from 6.6 to 8.1. The one score above 8 went to the one mechanism we called inventive without a qualifier. Scores in the sevens went to machines with a strength we could name without hedging, a multiplier bank that works as printed, a bonus that arrives every sixty-odd spins, a price list printed in full, and whose deductions leave that strength standing. Scores below 7 went to machines where the deductions outweigh the praise, two or more of them at once.

The rest of the scale is reserved as follows. Nine and above is for a mechanism of that order in a build that also prints its return and the odds behind its headline, and no game has combined the two yet. Five to 6.4 is for a machine that works as its rules say and gives us no strength to name. Below 5 is for a build whose rules leave out what a player needs to follow a spin. Below 3 is for a build whose behaviour contradicts its own rules, and only after a second session and a check of our own log have confirmed the contradiction. We have published nothing in those bands, and this page will say so when we do.

The tenths are ordinal. They place a game between its neighbours on the shelf, the same studio's first, and the review says which neighbours and why. A gap of a tenth or two means two games are level in our reading, and we do not claim that a second tester would land on the same tenth.

Two worked examples

Fortune Rabbit, 6.8. Modest credit for the mechanism, a count to five that is sound without being deep. Credit for legibility: five tiles pay the sum printed on them, in the base game and in the feature alike, and the player can watch it happen. Deduction for disclosure, the heaviest on the page: the build prints neither its 96.75% return nor its feature rate, the studio's sheet leaves its 0.77% feature row undefined, and nothing anywhere gives the weights of the tiles. The headline adds no second charge, because the 5,000x the loading screen sells is the cap the rules print and the odds it lacks are those same tile weights. Three logs stayed far from that cap, which the review reports as an observation. The base game neither adds nor costs much. The score lands below Speed Winner, 7.2, from the same studio, where every multiplier is visible before the win resolves, and a fraction above Royal Joker, 6.7, which pairs an unprinted return with one paying spin in ten.

Rabbit Garden, 7.3. Credit for disclosure: the return is printed on the fourth page of the in-game rules, where a player can read it before the first spin, and the price list is complete. Credit for legibility: the coin rule can be watched and checked as it runs. Deduction for the base game, the only deduction in the 7.3: most winning spins return less than the stake, so the table looks busy while the balance slides. The headline is not charged, because the 10,000x maximum is printed in the rules as a cap that ends the round, and a cap is a limit and not a promise. Its shortest route runs through a 10x tier that eleven logged rounds never visited, which the review reports and does not weigh. The rules give no weights for the coins and no odds for the climb. The review records that gap as a limit on what it can say, and the 7.3 carries no deduction for it. It sits above Fortune Rabbit for the return it prints, and below the eights for a base game that refunds more often than it pays.

How this section changes

This section was first published on 17 September 2026, after fifteen scores were already on the site, all of them set on the first five reasons. It binds the scores that follow, and on the same day it was turned on the fifteen it found. The closing sections of six older reviews were reworded where they had leaned on an event our logs never met or on a benchmark this site never defined. Then the base game of every portal session was measured one way: the share of paid spins that return anything, the share that return the stake or more, and the feature clock with its range. Each game was set beside its neighbours on the five reasons that scored it. Ten scores held. Five moved, by two to four tenths: Coin Princess x1000 to 7.2, Little Farm to 7.0, Thunder Coins to 6.9, 777 Coins to 6.7 and Floating Dragon: Year of the Snake to 6.7. Each move, and each closing section reworded without one, is logged in that review's Corrections list. When this section changes again, the change is dated here, and a game scored under the later text says so in its own Corrections list.

When we update a review

We revisit a published review when:

• the provider changes a mechanic (an RTP update or feature change);
• a factual error is found in the review;
• a follow-up session turns up new findings;
• a source is retracted or corrected.

We update the page's "last updated" date for substantive changes only - formatting tweaks don't trigger a date change.

A score changes when the facts it rests on change, or when a game is tested again and scored under a later version of How we score. Either way the closing section is rewritten and the reason is logged in the review's Corrections list, whether or not the number moved.

Who writes these reviews

Reviews are written by named testers, each with a public profile and a stated specialism. Every tester plays in demo, logs their own session data, and publishes it alongside the review.

• Erik Lindholm - founder and editor; slot maths, observed against stated RTP, and multi-session samples.
• Mia Stone - Hold & Win and fixed-jackpot specialist.
• Alex Reeves - casino analyst; mechanics, RTP verification and bankroll maths.

A note on names. All bylines on this site are permanent editorial pseudonyms. We work in a legally and professionally sensitive industry, so our testers do not publish their legal names or personal social profiles. Each pseudonym is used consistently by the same real person, the sessions attributed to a byline are conducted by that person, and the biographical statements on the author pages describe real experience. Avatars are illustrations, not photographs. Full details are on each author's page.

Independence

Nobody pays for our review scores, our session data, or what we write: scores come from testing, not partnerships.

Responsible gambling

A review is information for a decision, not encouragement to play. Slots carry a built-in house edge and are designed to lose money over time. Set a budget you can afford to lose, treat it as already spent, never chase losses, and stop when it stops being fun. 18+ (19+ in some regions). Support: BeGambleAware, GamCare, Gambling Therapy.