What RTP actually means
Return to player is the share of everything ever staked on a game that the game pays back, averaged over a number of plays far larger than anyone will ever make.
RTP is a property of a game's design, in the same way that a coin's fairness is a property of the coin. It is calculated from the paytable and the probability of each outcome, not measured from what players actually experienced.
The number is a limit. As the number of plays grows without bound, the share returned converges on the published figure. It says nothing about where you are on the way there, and for any realistic number of sessions you are nowhere near the limit.
A 96% RTP does not mean you get 96 cents back per dollar. It means that across all the money ever staked on that game by everyone, about 96% is returned as prizes. Almost none of it comes back to the individual player who staked it, and the distribution is deliberately lumpy: most plays return nothing, and the average is carried by outcomes that happen rarely.
The complement, 1 - RTP, is the house edge. A 96% RTP is a 4% edge. That 4% is the only part of the arrangement that is reliable as a long-run mean, and it applies to every coin staked — which is why the amount you stake, rather than the amount you deposit, is what determines the long-run cost.
Two figures are commonly confused with RTP. Hit frequency is how often a game pays anything at all, and it is unrelated to RTP: a game can pay often and small or rarely and large at the same RTP. Volatility describes how widely results spread around the average, and it is also independent of RTP.
To see what an RTP assumption does to a wagering requirement, use the expected balance calculator. To see how widely results scatter around it, use the variance explorer.