Why do my player props keep losing when I win about half of them?

You are not imagining it, and it is probably not variance.

Every page on this question says the same two things: bankroll management, and variance. Both are real and neither answers the question. The actual answer is arithmetic, and it is checkable. Half your props winning was never breakeven. It is a loss, and the size of that loss depends almost entirely on where you placed the bet.

A 50% record is not breakeven, and here is what it costs

A sportsbook prices both sides of a prop so the two implied probabilities add up to more than 100%. The amount over 100 is the overround. It is the whole of the book’s margin on that line, and you pay it whether you are right or wrong.

If the two sides are priced evenly around the line, an overround of V leaves each side asking you to be right (100 + V) / 200 of the time. A flat 50% record returns exactly V / (100 + V) of your stake, as a loss. That is not a model. It is the definition rearranged.

A bettor going 50/50 across a thousand $50 props is down somewhere between $2,565 and $5,115 depending on nothing but which app they opened. No bad beats required.

The range, measured on one night

The number above is only worth anything if somebody measured it, so we did.

We took our own board for the 12 September 2026 slate and kept every case where a single book quoted both the over and the under of the same player line. That is 6,213 player lines. For each one we converted both prices to implied probability and subtracted 100. That is the overround, and it is the only calculation the word actually refers to.

The result across the books on that board:

• NoVig was cheapest at 5.40%.
• Fliff was widest at 11.36%.
• Everything else sat between them.

Three things about that, because the honest version is more useful than the clean one.

It is one night. Book pricing moves with the slate. A single evening is not a season and we will not describe it as one.

We are not ranking the middle. The books between those two ends sit within about half a point of each other, and when we recomputed on only the markets every book quoted, some of them swapped places on moves smaller than the gap itself. A five book ladder would be reporting noise as a finding. The two ends held up under that test and got further apart, so those are the two we name.

Different books hang different markets. On that slate FanDuel priced seven markets and Fliff nineteen. Thin markets carry more margin, so a book that only offers main lines is flattered by a headline average. That is a real limit on any comparison of this kind, including ours.

We left prediction markets out of the range entirely. Polymarket’s figure on that board was far wider than any sportsbook’s, on a hundred or so rows, and it is a different mechanism being asked a different question. Putting it in a sentence about what sportsbooks charge would be wrong about what it is.

The book with the most to bet on is the most expensive place to bet it

This is the part worth keeping.

Fliff hangs both sides on more markets than any other book on our board, nineteen of them. It also has the widest overround we measured, 11.36%. The breadth and the price are not a coincidence in your favour. A long menu of unusual props is a long menu of thinly priced props, and the margin on them is where the breadth is paid for.

If you find yourself in an app because it is the only one carrying the market you want, that is the moment the number above is doing the most damage.

Where you bet moves more than which prop you pick

The overround describes one book’s margin on one line. The other question is what you give up by not checking a second book, and that is a different measurement: compare the two books’ prices on props they both quoted, row by row.

On the same night, 1,512 props that both DraftKings and FanDuel quoted:

• DraftKings was the better price on 70.5% of them.
• FanDuel was better on 25.9%, and 3.6% were level.
• The typical row differed by 1.26 points of implied probability.

Seven rows in ten, and more than a point of probability on the median row. That is larger than most of the edges people go looking for.

One caveat that bounds it: on that board DraftKings quoted 78.2% of the props and FanDuel 26.9%. The comparison only covers the rows where both were present, which is the correct basis for a head to head, but it means we are describing FanDuel on the quarter of the board it chose to price.

What this page is not telling you

It is not telling you the arithmetic can be reversed. Knowing the overround tells you the bar. It does not tell you that you clear it, and this page makes no claim about our own results: our pick record is being restarted as it goes public, so there is nothing here to cite yet and we would rather say that than point at an old number.

It is also not telling you to stop betting props. It is telling you that “I go about 50/50” is a description of a losing position, and that most of the difference between a 5% tax and a 10% tax is a decision you make before you look at a single stat.

Check it yourself

Any price, in ten seconds: put both sides of a prop into the no vig calculator and it will show you what the book is holding and what the two sides are worth without the margin. If the number comes back near 11, you have found an expensive line.

If you want the arithmetic behind why an even record loses, why -110 is not a coin flip is the concept with no measurement attached. This page is the measurement.

And the Lab board prices every prop we track against a fair built from the other books on that same line, so you can see what a market is charging before you take it rather than afterwards.

Numbers on this page come from our own board for the 12 September 2026 slate, pulled the evening before. One night. No sportsbook was contacted and there are no affiliate links on this page.