House edge and return are two views of the same cost
If a game has a published return to player of 96%, the implied house edge is 4%: 100% − 96%. Over a very large amount wagered under an unchanged model, the mathematical expectation is that 96 units are returned for every 100 wagered. Individual results can be far above or below that figure because variance operates over every small sample.
This calculation is useful only when the 96% statement applies to the exact game, mode and configuration being discussed. A site-wide “up to” figure is not enough to assign that return to a particular Rust case, crash game or battle.
Why game labels are not enough
Two pages can both say “coinflip” while charging different fees or using different item-value tolerances. Two crash games can use different formulas, maximum multipliers and edge adjustments. Every case can have its own distribution. The name identifies the format, not the price.
Our lower-edge comparison therefore requires a published percentage or a complete probability model. When neither is public, the edge remains unknown. We do not estimate it from animations, a short round history or promotional copy.
Calculating expected value from a case table
When a case publishes every item probability and credited value, the expected return can be calculated by multiplying each value by its probability and adding the results. Divide that expected return by the case price to obtain an implied return percentage.
For example, a 10-unit case with four outcomes might return values of 2, 5, 12 and 50 units with stated probabilities of 60%, 30%, 9% and 1%. The expected return is 2×0.60 + 5×0.30 + 12×0.09 + 50×0.01 = 4.28. That implies a 42.8% expected return under those example inputs. The example is intentionally illustrative; it is not a claim about a live case.
Player-versus-player fees
In jackpot, coinflip and battle formats, the platform may take a direct percentage, retain a portion of value or price items using its own internal balance. A nominally even winner selection can therefore still have negative expected value. Read the fee and valuation rules together.
Item liquidity creates another difference. A credited site value is not automatically the amount an item could be sold for on an external market, and the available withdrawal inventory may use different pricing again. The skin valuation guide explains those layers.
What fairness verification contributes
A verifier can help establish that a completed result follows committed inputs. It does not reveal the house edge unless the mapping from random input to outcome and the applicable payout table are also available. Technical integrity and price are related questions, but they are not the same question.
Put every percentage on the same denominator
Before placing two figures in one ranking, write what the percentage divides by. Case RTP commonly uses expected credited item value divided by case price. A coinflip fee may be a percentage of the combined pot. Rakeback can use turnover or theoretical loss. A skin spread compares two price observations. Those percentages are not interchangeable merely because they all look like costs.
| Published figure | Minimum evidence needed | Common mistake |
|---|---|---|
| Game RTP | Exact version, complete payouts and probabilities | Applying a site-wide “up to” number |
| House edge | Same model as the stated RTP | Estimating it from a losing session |
| P2P fee | Fee base and winner settlement | Ignoring unequal entry values |
| Case expected return | Every item chance and one valuation basis | Calculating only pictured prizes |
| Conversion spread | Same item, synchronized quotes and fees | Calling Steam Wallet value cash |
Turnover is not the same as money deposited
A ten-unit balance can be wagered repeatedly. If it turns over 20 times, total stakes equal 200 units even though only ten entered the account. A 4% edge applied to 200 units has an 8-unit mathematical expected cost under a fixed model. It would be incorrect to call that “4% of the deposit.”
Actual sessions will differ because balance may be exhausted early or grow, wager sizes change and game outcomes vary. Expected cost is an analytical average, not an invoice. For bonus play, contribution and restricted-balance rules add another denominator; use the wagering calculator before combining the figures.
Confidence labels make a comparison more honest
Use calculated only when a complete probability-and-payout model is available. Use operator-published when the exact game page states a percentage but the full model cannot be reproduced. Use partial when only one cost layer is visible, and unknown when the inputs do not support a number.
Do not upgrade an operator-published figure to independently verified because a verifier exists. Independent technical testing, when applicable, has a wider scope than reproducing one completed seed calculation. The UKGC testing strategy, for example, distinguishes rules, mathematical design, RNG output, scaling and implementation testing.
Comparable edge record
Only compare percentages tied to an exact game, mode, rule version and payout configuration.
| Field | What to record | Why it matters |
|---|---|---|
| Published return | RTP or complete outcome table | Establishes the mathematical starting point |
| Implied edge | 100% minus comparable return | Expresses long-run model cost |
| Additional fee | Rake, settlement deduction or game contribution | Captures costs outside headline RTP |
| Valuation layer | Deposit and withdrawal price basis | Shows skin-market friction around play |
Interpretation: Short histories cannot estimate a reliable edge. An unknown input should prevent a forced numeric comparison.
Before comparing any two games, record the precise version, rules URL, edge or fee, payout table, checking date and whether the figure is operator-published or independently assessed. If one field is missing, describe the gap instead of forcing a ranking.