Up or Down

What the house actually makes

Casinos do not usually publish this. There is no reason not to: the edge here is one number in a config file, it is printed on every bet you place, and knowing it does not help you beat it.

The edge is a spread, not an outcome

Each side of a round is sold at its chance plus a fixed spread and pays 1.00. The spread is the entire business. Nothing else about a round is tilted — not the candle, which belongs to Binance, and not the result, which is arithmetic on two published numbers.

At the default spread of , a coin-flip round prices both sides at 53¢:

side costs     0.53
pays           1.00
multiplier     1.00 / 0.53  =  1.8867×
your EV        0.50 × 1.8867 = 0.9434   →  −5.66% per bet
both sides     0.53 + 0.53  =  1.06 to win 1.00

So on a near-even round the house keeps about 5.7% of what is staked. Compare a European roulette wheel at 2.7%, an American one at 5.26%, a typical sportsbook at 4–5% on a two-way line. This is not a cheap market, and it is not pretending to be one.

A margin is not a win rate. Players win roughly half of all near-even bets here, by design — that is what a coin flip does. The house is not winning more rounds than you. It is being paid 1.8867× on something that happens 50% of the time and would need 2.00× to be fair.

What it costs against a player who plays well

5.66% is the margin on a round priced at even. Real rounds lean, and a player who always buys the side that is already ahead is buying favourites, where the same 3¢ is a smaller slice of a dearer share. So the honest test is to run that strategy against real data. On 12.5 days of real BTC candles, against the best strategy available to a player:

Every bet records the spread it was sold at, so these figures stay correct across a change.
Spread50/50 paysHouse marginBets to beat the noiseSides not offered
1.961×4.44%82214.3%
1.923×5.75%48917.0%
3¢ — default1.887×6.99%32819.4%
5¢ — the cap1.818×9.31%18123.3%

Past 5¢ the configuration refuses to start. At that point nearly a quarter of all sides would price above 99¢ and stop being sold at all — a book that will not quote a fifth of its own market is not a market.

The column nobody publishes

Bets to beat the noise. A single round moves the book by roughly half a stake, in either direction. The edge moves it by about six hundredths. So for the first few hundred bets the house's profit and loss is almost entirely luck, and says nothing at all about whether the pricing is right.

At 3¢ it takes around 328 bets before the edge is larger than one standard deviation of the noise around it. Below that number, a house that is down is not necessarily broken and a house that is up is not necessarily earning — both are ordinary variance, and anyone reading a short P&L as a verdict on the model is reading static.

This cuts the player's way too, and harder: the same arithmetic run from your side is why a good session proves nothing about a strategy. Several hundred bets is where the coin stops being louder than the edge — and by then the edge has been quietly charged on every one of them.

What is not in the price

Why say any of this

Because the alternative is that you assume something worse. A market that will not tell you its spread is asking to be trusted about the one number that decides whether you can win, and the whole design here runs the other way: the mechanism is written down and the result is somebody else's to publish. The edge is the last piece, and hiding it would make the other two decorative.

Open the live market