A payout can be quoted as profit relative to the stake or as the total amount returned including the stake. Confusing these conventions changes the arithmetic by one stake. Roulette provides a simple example because a familiar straight-up payout can be evaluated under an explicitly stated wheel model.
Assume an ideal single-zero wheel with thirty-seven equally likely pockets and a straight-up bet paying thirty-five to one in profit. A winning stake of one earns thirty-five profit and returns the original stake as well. Total winning return is therefore thirty-six, not thirty-five. A losing selection returns zero in this simple model.
The expected return per unit staked is thirty-six divided by thirty-seven, approximately 0.9730. Expressed as a percentage, that is about 97.30%. The complementary house edge is roughly 2.70%. This is model arithmetic over repeated play, not a prediction of the next spin or a guarantee for an individual session.
Change the model to thirty-eight equally likely pockets while retaining the same straight-up payout. Expected return becomes thirty-six divided by thirty-eight, about 94.74%, and the house edge is about 5.26%. The extra losing possibility changes the comparison even though the payout label is unchanged.
Those examples do not describe every roulette product. Special payouts, zero treatment on even-money selections, side bets and other rules can require different calculations. The exact wheel and paytable must be identified before assigning any figure to a real game. A generic roulette category page is not enough evidence.
Bonus terms are another separate layer. A casino may exclude a roulette variant or give table-game stakes a different contribution towards wagering. A mathematical return estimate for the game does not establish promotion eligibility or the conditions for a withdrawal.
When reading a comparison, ask whether return includes the original stake and whether the listed outcomes cover the complete model. Then ask which actual game version the assumptions represent. This approach keeps an understandable example useful without turning it into an unsupported claim that one operator’s entire casino has a particular return or that a progression of stakes can remove the underlying edge. Keep the assumed pocket count and payout convention visible so the reader can reproduce the calculation.