What is a middle in sports betting?

Most books have an NFL favorite at -3. One has moved to the favorite at -2.5, priced at -110; another still has the underdog at +3.5, also at -110. A gap like this closes as soon as the slower book moves. Bet $110 on each side:

Only one result, the favorite by exactly 3, wins both bets.

A miss loses $10 of the $220 staked, or 4.55%: the books' margin on the pair. ArbWins sizes the stakes so a miss costs the same either way: each leg's share is the other leg's decimal price over the sum of both, an even split at equal prices.

The arbitrage calculator gives the same split. Enter the two prices; when they add up to more than 100%, the negative edge it shows is exactly what the pair costs on a miss. If a line is a whole number, a result on that number pushes that ticket and refunds it instead of losing it. With half-point lines, as here, there are no pushes.

A middle is worth taking when the chance of landing in the gap beats its break-even hit rate: what a miss costs, divided by what a hit wins plus what a miss costs. Here that is $10 ÷ ($200 + $10) = 4.76%. The prices matter as much as the gap. Around key numbers, books often charge more than -110, and the cost climbs fast:

The second pair costs 8.33% on a miss, over the 8% limit on the middles board, so it would not be listed at all.

NFL points mostly come in 3s and 7s: a field goal, and a touchdown with the extra point. Final margins bunch on those numbers. Since 2003, about 15% of NFL games have been decided by exactly 3 points, more than any other margin, and 7 comes next (Action Network's margin data). A window that covers 3 is worth far more than a window of the same width that covers 4 or 5.

That 15% counts either team winning by 3. Only the favorite winning by 3 hits here, and with the line near 3 the favorite wins the larger share of those games. So the realistic hit rate is below 15% but comfortably above the 4.76% the -110 pair needs; at -120, needing 9.09%, there is far less room for error. A -6.5 and +7.5 pair is the same idea on 7. NFL totals and basketball results have key numbers too, but they are far weaker.

Each row on the middles board shows two numbers side by side: Need to hit, the exact break-even rate, and Model est., ArbWins's estimate of the chance of landing in the window.

The model treats the final margin, or the total, as a normal distribution centred on Pinnacle's line for that game, with a fixed spread per sport: a standard deviation of 13.5 points for NFL margins and totals, 11.5 for NBA margins, 18 for NBA totals, and about 2.2 goals for NHL margins. It counts the whole-number results strictly inside the window, with half a point of padding either side because scores are whole numbers. When Pinnacle has no line for the game, it centres on the midpoint of the two lines. Player props work the same way, with a spread for each stat.

For the example above, with Pinnacle at 3, the model puts the favorite winning by exactly 3 at 2.95%. That is below the 4.76% needed, so the row would show an expected value of about -1.7% of the stake. This is the model's known weakness: a smooth curve gives every whole number near the line about the same share, so it cannot see that NFL games land on 3 and 7 far more often than on the numbers around them. On a window over 3 or 7, the real hit rate has historically been well above the estimate, so on an NFL window, check which numbers it covers.

Both legs have to settle the same way. At the books ArbWins reads directly, NFL spreads and totals include overtime, and the board never pairs a full-game line with a regulation-time line. The settlement rules table has the detail for each sport and book.

Sportsbooks limit middle bettors for the same reason they limit arbitrage bettors: the bets land on lines the book has left behind. And a middle loses most of the time by design, so it needs the patience of +EV betting. Compare arbitrage betting, which pays whichever side wins, and ArbWins's track record of how long gaps between books stay open.