Theoretical and realised RTP: an average is not an individual promise

Define the theoretical quantity

A theoretical return ratio comes from the probability model and its output rules. The Gambling Commission’s public explanation describes RTP as an average over many plays rather than an amount guaranteed each time. That source concerns Great Britain and is used here for terminology, not to claim licensing or approval for this website. An average does not specify every possible outcome, the spread of outcomes or a particular person’s result. A complete model needs the distribution, not merely the single number printed as a percentage.

Use an abstract output model

Suppose one hypothetical input unit produces output 0 with probability 0.50, output 1 with probability 0.25 and output 2.8 with probability 0.25. The expected output is 0×0.50+1×0.25+2.8×0.25=0.95 units. These invented values are a teaching device, not the rules of a real product. OpenStax’s expectation formula supplies the weighted-average calculation. The individual outputs remain 0, 1 or 2.8; there is no rule requiring a particular input to return 0.95.

Inspect contributions and variability

The variance is 2.21−0.95²=1.3075 output units squared. This derived quantity shows that the same mean can coexist with substantial variation. A different distribution could share the 0.95 expectation while producing a different variance. Therefore an expectation alone cannot reveal the range or concentration of possible outcomes.

Calculate a realised ratio from a complete record

For a specified observation window, a realised ratio divides aggregate recorded output by aggregate input. The Commission’s monitoring explanation uses that aggregate distinction and discusses variability. In a hypothetical record of ten equal input units with total output eight, the realised ratio is 8/10=80%. This is an observed summary of the invented record, not a revised theoretical parameter. Define the window, included transactions and data treatment before calculating. Excluding zero-output observations or selecting only successful periods would distort the ratio.

Avoid the false promise of eventual recovery

A gap between a realised ratio and the theoretical expectation does not create a debt that later observations must repay. Under an independent fixed model, the next output distribution remains the same. More observations can reduce relative variability of an average under suitable assumptions while adding further exposure to the process. A theorem about averages does not promise that an individual cumulative result returns to a chosen target. The historical total is already part of the record and cannot be changed by renaming the next observation as a correction.

Do not average incompatible percentages

If two observation windows have different total inputs, their realised percentages should not simply be averaged with equal weight. For an abstract example, one window has output 9 from input 10 and another output 80 from input 100. The combined ratio is 89/110, approximately 80.91%, rather than the unweighted average of 90% and 80%, which is 85%. This is a general property of ratios. Preserve the raw numerator and denominator so aggregation remains correct when periods or datasets are combined.

Separate monitoring from individual predictions

A monitoring analysis examines whether aggregate observations are compatible with a specified model and data process. It may need a variance estimate, a planned observation window and treatment of repeated checks. A single departure from an expectation does not automatically identify its cause, and compatibility does not certify every aspect of a system. This article does not audit a product or assert that any named domain has passed testing. Its narrow purpose is to prevent a mathematical average from being presented as an individual entitlement or a method of selecting future outcomes.

RTP interpretation checklist

• Distinguish theoretical expectation from realised ratio.
• Read the complete output distribution when available.
• Keep numerator, denominator and observation window.
• Weight combined ratios by their denominators.
• State independence and stability assumptions.
• Reject claims of guaranteed compensation or recovery.
• Do not infer licensing, safety or endorsement from an average.

When explaining the concept publicly, place the limitation beside the percentage rather than in a distant footnote. A reader should understand immediately that the figure describes an aggregate model. Use abstract, clearly labelled examples and avoid turning the explanation into a comparison intended to encourage participation.

Sources and verification

• Return to player: interpreting the average — Gambling Commission, Great Britain — 2026-10-23
• Mean, expected value and standard deviation — OpenStax, Rice University — 2026-10-23
• Calculation and monitoring of actual RTP — Gambling Commission, Great Britain — 2026-10-23