Observed RTP: what a sample can tell you

A slot’s published return to player (RTP) is the share of all stakes it is designed to pay back over its lifetime. The tracker’s observed RTP is the share of stakes that the bets it saw in casinos’ public feeds paid back over the last 30 days, to 6 Nov 2026, 15:12 UTC. This post explains the difference, and how to read the 95% interval beside each figure on the observed RTP page, using the most-played slots and the casinos’ own games.

Published RTP is a long-run average

Test labs define RTP as an expectation. Gaming Laboratories International (GLI) calls it “the expected percentage of wagers that a specific game will return to the player in the long run”. eCOGRA describes the return expected “over an extended period”. Both derive it from the game’s maths, by calculation or simulation.

The house edge is the casino’s side of the same number: a 99% RTP means a 1% edge. As an example, $1,000 staked at a 1% edge is expected to cost $10 in the long run, though one session can end far above or below that.

Regulators and labs set rules around it. The UK Gambling Commission’s technical standard RTS 3 requires information such as the house edge or RTP to be “easily available before the customer commits to gamble”. Malta’s Player Protection Directive requires licensed online games of chance to pay out on average 85% or more of stakes, measured by default over “a period of one calendar year”. GLI’s online standard, GLI-19, in a version dated 3 November 2026, says that, unless a game’s rules provide otherwise, “a game shall not adapt its RTP based on past payouts”. It also has labs test random number generators for “statistical independence between draws”, so a run of losses does not make a win more likely.

Observed RTP is a sample

The tracker’s observed RTP is total paid divided by total wagered, over the bets it saw. The UK Gambling Commission’s guide to live RTP monitoring applies the same sum to all of a game’s play and calls the result the “actual RTP”. In its example, a game designed to return 91.68% takes £1,200,000 in stakes in a month and pays out £1,085,000: an actual RTP of 90.42%, “below the designed RTP”. Whether that gap matters, the guide says, depends on the game’s volatility and on how much play was measured.

The tracker’s figures differ in two ways. Public feeds show some bets, not all. And tracking began in November 2026, so the 30-day window so far holds only the days since then.

How the 95% interval works

The interval is the range that chance alone could produce around the observed figure. If a game pays exactly as designed, its published RTP should fall inside the interval about 19 times in 20. The width depends mainly on the number of bets and on volatility, meaning how unevenly a game pays. A high-volatility game tends toward prizes the UK Gambling Commission calls “very large but rare”, sometimes at a big multiplier (the payout divided by the stake), as the post on slot volatility explains.

The UK Gambling Commission’s guide shows how fast the range narrows. For its example game, with a standard deviation (a measure of spread) of 5.6, the 95% range is about 4.9 percentage points either side of the designed RTP after 50,000 games, 3.5 after 100,000, 1.7 after 400,000 and 1.1 after 1,000,000. Four times the play halves the range. Even then, “a non defective game might still fall outside range approximately 1 in 20 tests”.

Labs test against the same kind of range. GLI’s field data checks, done on request, report whether live payouts are “within statistically acceptable bounds”. eCOGRA’s tests, run on large simulations, determine whether a game’s payout “falls within a statistically acceptable interval”. GLI-19 adds that where a game displays an actual RTP, it must state the number of plays and the period, as every figure below does.

The most-played slots, last 30 days

These are the most-played slots on the observed RTP page, over the 30 days to 6 Nov 2026, 15:12 UTC.

Read the intervals, not the middle column. Gates of Olympus 2500 has the most bets and still spans 81.81% to 112.06%. Sweet Bonanza 2500 reads 90.00%, but its interval reaches 102.48% and holds the published 96.52%. Deadline Max Chaos, at 135.13% on a range of 32.88% to 237.38%, says little either way.

The Dog House is the exception. Its whole interval, 99.47% to 149.34%, sits above the published 96.51%: in this sample, it paid back more than its design implies. A game working as designed lands outside its range about 1 time in 20, and the UK guide says one such failure “does not confirm the game/RNG is faulty”. Every other slot here has its published RTP inside the interval, though Big Bass Splash only just: its range starts at 96.19%, against a published 96.71%.

Why big bets widen the range

Bet sizes matter too, because observed RTP weights each bet by its stake. As an example, one $1,000 bet counts as much as a thousand $1 bets. So when stakes are uneven, a few large results can move the average, and the interval widens. On Sweet Bonanza 2500, $126.9 million was wagered over 114,620 bets in the 30 days, and single bets get very large: the high-roller list shows three of $1,002,000 each on it at Stake between 12:52 and 12:55 UTC on 3 Nov 2026. The UK guide warns that play at maximum stakes “might drown out activity played at minimum bet levels”.

Originals: bigger samples, mostly narrower ranges

Originals are games casinos offer under their own name, such as dice, limbo, plinko, mines and keno. Most of those below advertise a 99.00% RTP, a 1.00% house edge. GLI-19 lists dice, roulette, keno, crash and mine games as house-banked games, like slots, where the casino takes the other side of every bet. These are the most-played originals with an advertised RTP on the originals RTP page, over the same 30 days.

Every advertised figure sits inside its interval, some only just. Dice at Housebets reads 97.45% with an upper bound of 99.96%, and limbo at Gamdom reads 92.82% with an upper bound of 99.24%. Neither sample can tell its game apart from one paying exactly as advertised.

Most of these ranges are narrower than the slots’, helped by larger samples. The narrowest in this data belongs to hilo at CSGOFAST, which the tracker shows without an advertised figure: 95.98% over 1,198,469 bets, with an interval of 95.61% to 96.36%. Keno at Gamdom stays wide despite the most bets in the table, because its results swing more from bet to bet.

Crash at Stake: counting rounds

In crash, a multiplier climbs from 1.00× until the round crashes, at a point GLI-19 requires to be “randomly determined”, and a bet pays only if it was cashed out first. For Stake’s crash game, the tracker measures the edge from rounds rather than bets, so bet sizes don’t weigh on it. As of 6 Nov 2026, 15:12 UTC it had recorded 10,948 rounds, with an observed house edge of 1.14% (95% interval 0.81% to 1.47%) against an advertised 1.00%, which sits inside the interval. Rounds reached 2× in 49.18% of cases, where a 1.00% edge predicts 49.50%.

What the numbers can’t show

These figures describe the bets the tracker saw, not everything that was played. The interval measures chance within those bets; it cannot correct for what a feed leaves out. The totals pool many players, and one person’s session is a far smaller sample; smaller samples swing more. With many games on the board, a few can be expected to sit outside their range by chance alone.

None of this changes the house edge, which means players as a group lose money over time; big wins are rare by design. Gambling is for adults (18+). The responsible gambling page covers limits, warning signs, self-exclusion and help lines.

What to watch next: as the 30-day window fills, the ranges on Sweet Bonanza 2500 and Deadline Max Chaos should narrow if bets keep coming in, and The Dog House will show whether its reading drifts back toward its published 96.51%.