We Recorded The Same Slot Twice And Got 94% RTP, Then 119%

We record real slot sessions and publish what we see, including when the numbers look strange. This is one of those times. We recorded Hell Butcher twice, about 1,200 spins each time, and got two very different answers. Session one returned 94.3%. Session two, on the same slot at the same stake, returned 118.9%. A 24.6 point spread on a game with a published theoretical RTP of 96.55%.

That alone would be a fun coincidence. What makes it useful is what happened on a different slot. Flames and Greed ran 3,267 spins in a single recorded session, nearly three times the length of either Hell Butcher session, and came in at 57.1% observed RTP against a theoretical 96.12%. More spins did not bring the number closer to theoretical. It went the other way.

Three recorded sessions is a small sample and we are not going to dress it up as more than that. But three sessions is enough to make a point that gets ignored constantly in slot discussion online: a single session's RTP, yours or a streamer's, tells you almost nothing about a game's real return.

What We Actually Recorded

Each session below comes from a real play recording, not a simulation. We log every spin's stake and payout, tag bonus rounds separately from base game spins, and compute an observed RTP and hit frequency from the raw spin log. Every session also gets a data quality score based on how cleanly the recording captured each spin.

Only sessions that clear two conditions get their per spin numbers published as findings. Data quality has to be at or above 85%, and the session has to be flagged reliable at the per spin level. Below that bar, a recording can still contain real spins, but small gaps or misreads accumulate into numbers that no longer mean anything. We do not average a shaky session in with a clean one and we do not footnote the problem away. We hide the RTP figures entirely.

Three sessions across two games clear that bar right now. Here they are.

Hell Butcher carries a theoretical RTP of 96.55% at medium volatility. Flames and Greed sits at 96.12% theoretical and runs high volatility. Neither observed figure above should be read as a corrected RTP for either game. They are exactly what the table says: what these specific recorded sessions returned.

Why Did Hell Butcher Return 94% One Session And 119% The Next

Same slot, same stake, roughly the same spin count, and a 24.6 point gap. Hit frequency landed identical at 31.5% in both sessions, so the difference is not about how often the slot paid. It is about how much the wins were worth when they hit. Session B's biggest win was 516.6x stake against session A's 245.1x, and session B also produced two more bonus rounds. A small handful of results at the top of the paytable moved the whole session's return by double digits.

That is the ordinary behavior of a slot with a real paytable, not a flaw in the recording. At 1,200 spins the sample has not run long enough for the rare, large outcomes to average out. Whether a session lands on the high or low side of a few big spins decides the headline number almost by itself.

Why More Spins Did Not Get Flames and Greed Closer To Theoretical

Flames and Greed's 3,267 spin session is the more interesting result of the two, because the natural assumption is that more spins means a more accurate number. Here it ran roughly three times longer than either Hell Butcher session and still finished 39 points under its own theoretical RTP, at 57.1% against 96.12%.

The reason comes down to volatility. Flames and Greed runs high volatility against Hell Butcher's medium, which means more of its long run return is concentrated in rarer, larger outcomes. A high volatility game's realized return converges toward its theoretical RTP far more slowly than a medium or low volatility game's does, because the sample has to be large enough to actually capture those rare outcomes at their true frequency. Twenty seven bonus rounds triggered in this session, but that is a count of rounds, not a statement about what any one of them paid. We are not attributing this session's shortfall, or Hell Butcher's high session, to any specific feature. The data only supports a base game versus bonus round split, nothing more granular than that.

Put plainly, spin count is not a substitute for understanding a game's volatility profile. A thousand spins on a low volatility slot will sit much closer to theoretical than three thousand spins on a high volatility one.

What We Throw Away, And Why That Matters More Than What We Publish

We have recorded more than three sessions. Four others exist right now that do not appear in the table above, because they failed our quality bar. We are naming them here instead of quietly leaving them out, because the exclusion is the actual point of this article.

Necro Party has a recorded session of 5,484 spins, longer than any of the sessions above, and its data quality score came in at 38.4%. That session's raw numbers once rendered live on this site as an observed RTP of 830.1%, a figure that is obviously not real for any slot. We caught it, and the fix was not a caveat or a footnote. A dataQualityPct below 85%, or a session flagged unreliable at the per spin level, now suppresses the RTP and bonus share figures entirely rather than showing a number we cannot stand behind. Wild Skullz has three recorded sessions, at 84.6%, 55.4%, and 41.6% data quality, and none of them clear the reliability bar either. None of those five sessions' RTP figures appear anywhere on this site as findings.

This is the part of first party data work that matters most and gets talked about least. Recording real sessions is not hard. Knowing which recordings you can actually trust, and being willing to throw away the ones that produce numbers you cannot defend, is the part that makes the sessions you do publish worth reading.

What This Means If You Are Judging A Slot By One Session

If you have ever watched a streamer run bad on a slot for twenty minutes and concluded the game is rigged, or run hot yourself and told a friend a slot is "loose," the numbers above are the mechanical explanation for why that read is unreliable. A session of a thousand or even a few thousand spins sits well inside the range of normal variance for almost any slot with real volatility. Two sessions on the exact same game, recorded under our own quality bar, landed 24.6 points apart. That is not an edge case. That is what short run variance looks like on ordinary hardware.

None of this is an argument to play more than you intend to or to chase a number back toward theoretical. If anything it argues the opposite. A losing session is not evidence a slot "owes" you a correction, and a winning one is not evidence you have found an edge. Set a budget for the session, stick to it, and treat any single session's outcome, yours or someone else's on a screen, as one data point rather than a verdict on the game.

We publish observed data because raw numbers from real sessions are more useful than we could get from marketing copy alone, provided the numbers are labeled honestly. That is the entire premise of this article. Read more about how we source and verify data across the site on our E-E-A-T and methodology page, or see the live observed play data for Hell Butcher and Flames and Greed on their slot pages.