Spunk Odds Reference

Dice guide

Cosmos Dice: roll over, roll under, same odds

This guide carries the name Cosmos for the site's own filing, not because the game differs. It looks at the two directions of a number-dice bet and shows that, counted carefully, they are mirror images. The example uses a 1.5% edge.

Two ways to describe the same slice

Picture the 10,000 possible results laid out in a line, from 0.00 on the left to 99.99 on the right. A roll-under bet claims a slice on the left. A roll-over bet claims a slice on the right. If both slices hold the same number of results, they have the same chance and, in this model, the same payout.

For a roll-under target t, the winning results are 0.00 up to t - 0.01, which is t x 100 results. For a roll-over target t, the winning results are t + 0.01 up to 99.99, which is (99.99 - t) x 100 results. To mirror an under bet with chance c%, set the over target to 99.99 - c.

Mirror pairs at a 1.5% edge: each row has the same chance and payout in both directions
Win chanceUnder targetOver targetWinning resultsPayout
12%under 12.00over 87.991,2008.2083x
37%under 37.00over 62.993,7002.6622x
50%under 50.00over 49.995,0001.9700x
71%under 71.00over 28.997,1001.3873x

Model: 10,000 equally likely results, payout (1 - 0.015) x 100 / chance.

Where off-by-one mistakes happen

The tricky part is the boundary. Is the target itself a win or a loss? In the convention used here, an under bet on 37.00 loses if the roll is exactly 37.00, and an over bet on 62.99 loses if the roll is exactly 62.99. Some services draw the line the other way and count the boundary as a win for the over side. That moves the chance by one result, 0.01%, which changes the payout in the fourth decimal.

If you compare two services, read their rules for the boundary first. An over target of 63.00 under this convention wins on 3,699 results, not 3,700. That is a 0.027% relative difference in chance, small but real, and it is the kind of detail a payout table should make explicit.

Why direction never matters for the average

Since the payout formula only looks at the count of winning results, the expected value is -1.50% per unit at this edge for both directions, and for every target. Switching from under to over after a run of high rolls changes nothing. The next result is still one draw from the same 10,000 equally likely values.

Under-bet view of four chances at a 1.5% edge; the mirrored over bets give identical rows
Win chanceRoll neededPayoutEV per unitSD per bet
12%under 12.008.2083x-1.50%2.667
37%under 37.002.6622x-1.50%1.285
63%under 63.001.5635x-1.50%0.755
88%under 88.001.1193x-1.50%0.364