Two slots can both print 96% and still empty a balance on different clocks. Return to player is a long-run share of stakes. Volatility is the shape of the path that gets there. A player who treats those words as synonyms will size the stake for the wrong game.

A useful check is a table whose rules sit on the cloth, where the price of a hand is visible before the cards move. Duel Blackjack states blackjack pays 3 to 2, the dealer stands on 17, and insurance pays 2 to 1. A slot rarely puts an equivalent sentence on the felt. The help file has to do that work, and most players open it after the balance has already moved.

What 96% does and does not say

A 96% setting means that, across a very large number of spins at that setting, the game keeps about 4 units from every 100 staked. It does not mean a 100-spin session returns 96. It does not mean the next bonus pays. It does not mean every casino loads that setting. Studios often certify a ladder. The figure in your session is the one the operator selected, which is why a serious note names the build, the date, and the screen, not a memory of a trailer.

Hit frequency sits beside return and is still a different number. A game can pay often in small amounts and still hold 4%. Another can go quiet for long stretches and then pay a single large multiple. Both can share a headline return. The quiet game needs a larger reserve per spin, because the gap between pays is the thing that ends sessions.

A simple map of session shape

These rows are shapes, not certified maths for a named title. Use them to pick a stake unit before you pick a theme.

Session shape	What you feel	Reserve that survives a dry run	Typical mistake
Frequent small pays	The balance ticks	80 to 150 stakes	Raising the stake because the game "feels safe"
Mixed, with a real bonus	Long base game, then a feature	150 to 300 stakes	Quitting the feature rules unread
Rare, heavy pays	Many dead spins	300 stakes or more	Treating a 50-spin sample as proof the game is cold
Bonus buy	You skip the base game and pay a multiple	The buy price, counted as one decision	Comparing the buy to a single spin instead of to its price

The third row is where bankrolls go. A high multiple on the paytable is not income. It is a rare cell. If your stop-loss is 40 stakes and the feature that carries the return shows up a few times per hundred spins, you will often finish at the stop without ever seeing the cell the trailer showed.

Stake unit, then theme

Pick the unit from the shape, then see whether the theme still interests you. Reversing that order is how a 2-unit stake lands on a game built for a 0.2-unit stake. Write three numbers before the first spin: the unit, the loss stop, and the time stop. A win stop belongs there too. Open-ended sessions turn a good feature into a second deposit.

Demo mode is the clean place to learn the paytable, the feature rules, and whether a buy exists. It is a poor place to judge return. Demo stakes are not a sample of the certified maths, and a short demo cannot show a rare max win. Use it to read. Use the real session, if you play one, at the unit you wrote down.

What to record if you compare games

Keep a line per title: studio, stated return if you saw it, whether a ladder was mentioned, volatility word or number, max win multiple, buy price if any, and your unit. Without that line, the next session borrows a feeling from the last one. Feelings are not settings.

A 200-spin sketch, still not a forecast

Take a 96% game and 200 spins at 1 unit. The theoretical hold is about 8 units. The session can finish ahead, behind, or near that mark. The spread is the volatility. On a frequent-pay game the path wobbles inside a narrower band. On a rare-pay game the path can sit 40 or 80 units down and still be inside a normal night, then jump on one feature. Planning as if every 200 spins owes you 192 back will make a normal dry run feel like a broken game.

Count features, not only the end balance. If the panel says the feature is the main source of the return, a session with no feature is an unfinished look at the game, not a verdict on the setting. That is also why a 30-spin "test" cannot rank two titles. You measured the start of the path.

Hit rate is not generosity

A hit rate of 1 in 4 means something lands often. It does not mean the landings cover the stake. Many hits pay less than the bet. The balance still trends down between features. Players call this "the game gives something back" and raise the stake. The hold does not care about that feeling. If you want a calmer balance, choose a shape from the table above and cut the unit. Do not promote a frequent-pay game into a high stake because the reels move.

Max win deserves one sober line in the same note. A 10,000x cell can be certified and still be so rare that a private player should not budget for it. Budget for the feature you can actually reach inside your stop. Treat the poster multiple as a ceiling the maths allows, not as a target for the evening.

When two games share a studio

Studios reuse maths shells. A new skin can carry the same return, the same hit rate, and a different feature story. It can also carry a quieter rung. The art will not tell you which. Open the panel on each title even when the cabinets look like cousins. Log them as separate rows. If you only log the brand, you will stake the second game at the size you used for the first, including the one time the operator loaded a different build.

Names the panel uses, and what they leave out

Volatility labels are not a shared scale. One studio's "high" is a number out of five. Another studio's "high" is a word with no number. A third game prints neither and lets the max-win multiple do the talking. You cannot rank those three on the adjective. Rank them on the multiple, the hit rate if it is printed, and the buy price if a buy exists. If only the word is printed, write the word and do not pretend you saw a figure.

Return and volatility can move independently across a ladder. A quieter rung is sometimes the same feature with smaller prizes, and sometimes a different feature weight. The panel will not always say which. When it is silent, your stake decision should follow the harsher reading: assume the dry spells can be as long as the louder build, and size the unit for that. You can always stake less than the game allows.

Demo play still helps here, with a limit. Use it to see how a feature ends, whether wins are capped inside the bonus, and how long a dead stretch feels at your intended speed. Do not use twenty demo spins to declare the live build "tight." You did not measure the certified return. You rehearsed the rules.

Tournaments and drops add a third layer. A prize pool can make a low-return spin rational for a player who is farming points, and a trap for a player who thought the pool replaced the hold. Read whether slot spins count, whether a minimum stake applies, and whether the prize is cash or bonus funds with wagering. Write that next to the unit. A leaderboard is not a paytable.