Pokies analytics · independent · Australia
Lucky Radar is a small analytics desk that takes poker machine mathematics seriously. We publish no offers, link to no operators and run no affiliate arrangements. What we do is show the arithmetic that machine specifications imply — and let you run it yourself below.
Return-to-player figures quoted by manufacturers are computed over tens of millions of simulated spins — not over an evening.
A one-hour session at usual speeds is a few hundred spins. At that scale variance, not RTP, decides what happens.
That is the whole formula. A 90% machine keeps 10c of every dollar cycled, on average, over the long run.
This is a Monte Carlo simulation, not a game and not a machine. It takes a return-to-player figure, a hit frequency and a session length, runs the session a thousand times, and shows you the spread of outcomes. Nothing is wagered, nothing is won — it is a histogram.
Press run. The bar chart shows how 1000 identical sessions ended, in credits.
Return-to-player is the proportion of everything cycled through a machine that the machine returns across its full theoretical cycle. Two points follow from that definition and both are routinely lost:
It is computed on turnover, not on deposits. A hundred dollars cycled twenty times is two thousand dollars of turnover, and the percentage applies to the two thousand. This is why "I got back 90% of my money" and "the machine returns 90%" are different sentences.
It is an average over the whole cycle. The figure is a limit, approached over millions of outcomes. Nothing obliges any particular thousand spins to look like it, and in practice they do not.
| Component | What it contributes | Share of return |
|---|---|---|
| Small frequent outcomes | Keeps a session alive; barely moves the total | low |
| Mid-range outcomes | The bulk of visible returns | moderate |
| Rare top outcomes | Carries a large share of the advertised figure | high |
Shares are qualitative on purpose. Exact splits are specific to each machine's paytable, and we do not publish numbers we have not computed from a published paytable ourselves.
Take the simulator above at its defaults: a 90% return, a quarter of spins returning something, three hundred spins. The average result across a thousand sessions lands near the arithmetic expectation — minus ten per cent of turnover. The median session lands somewhere else entirely, and the distribution is skewed, not symmetric.
That asymmetry is the point. A symmetric distribution would put half of sessions either side of the average. A skewed one puts most sessions below it, and pays for that with a thin tail of large results. Raise the return figure and the tail gets thicker; it does not make the typical session typical.
The median tells you what a normal session looks like. The average tells you what the machine's specification implies. When those two numbers disagree — and here they always do — quoting only one of them is how the category gets misunderstood.