Middle Calculator

Price a middle on both sides of a moved line: what it pays, what a miss costs, and how often it has to land.

This prices a middle you can already see and says how often that window has landed in the NFL, which is the frequency half of the same comparison; how often it will land in your game is still yours to judge.

Break-even hit rate

4.8%of the time, to cover the $10.00 a miss costs

This wins both bets if the result lands on 45 or 46, and one bet the rest of the time.

Over 44.5 wins→←Under 46.5 wins44or lower4546both win47or higher

44.546.5

Only the under wins−$10.00Both bets win+$200.00Only the over wins−$10.00

Where the break-even rate comes from

A miss costs$10.00Swing from that miss to a hit$210.00Break-even hit rate4.8%

Your two bets

Enter the two numbers you can bet, one on each side of the market. For a total, that is the points, goals or runs line; for a spread, the winning margin.

Number

Stake$110.00

Number

Stake$110.00

Total stake

Split so both one-sided results cost the same.

What each result pays

Result	Outcome	Profit
44 or lower	Over loses, under wins	−$10.00
45 or 46	Both bets win	+$200.00
47 or higher	Over wins, under loses	−$10.00

Total staked: $220.00
Both bets win: +$200.00
Each miss: −$10.00
Combined book: 104.76%
Gap: 2 points

How often it lands

What did this market close at? (optional)

For a spread, enter the home team's number: positive if the home team was favored (a home favorite of 3 is 3), negative if the home team was the underdog (an underdog of 3 is −3). For a total, enter it unsigned, like 44.5. Checked against 6,719 NFL regular-season games, 2000 to 2025, conditioned on the closing number rather than taken across every game.

How often you think it lands there5.0%

Expected profit per attempt, at your estimate

+$0.50break-even is 4.8%

Your estimate5.0%

Break-even hit rate4.8%

That percentage is the judgement the position turns on, and it is the one figure most middle calculators leave blank. A number invented here would be worse than none, so the two rows below are measured instead: how often these exact results have come up in 6,720 NFL regular-season games, 2000 to 2025. Your two numbers are read once as a game total and once as a winning margin, because nothing on the page can tell which market you are in and only one of the two is yours. Sourced in the methodology under this tool.

As a game total, landed on 45 or 46: 305 of 6,720 games (4.5%)
As a winning margin, either side won by 45 or 46: 10 of 6,720 games (0.1%); a ceiling, not an estimate

The margin row counts both teams, so it is the share of games decided by that many points either way. A spread middle needs one named side to win by it, and how that splits between favorite and underdog is not in this data, which is why the row carries no button: treat it as the most the window can be worth, and note that a ceiling already below your break-even rate settles the question without the split.

Both figures are also unconditional, taken across every game rather than across games the market hung at these numbers, and this site's own study of the same games measured what conditioning on the closing number actually does to each. For a total it barely moves the figure at all: the tightest total window reads 2.82% unconditional against 2.88% conditioned on its own closing number, about a twentieth of a point. For a spread it is a real difference: the tightest spread window reads 4.02% unconditional against 4.80% conditioned on its own closing number. Enter the closing number above for the reading built for this game, where the study reported one, rather than the average across every game in the sample.

Where the window sits, at one width and one price

Every row below is the same bet: a two-point totals middle, two sides at -110, break-even at 4.8%. All that changes is which two numbers the gap covers, and football scoring is lumpy enough that this moves the frequency across the break-even rate rather than around it.

Over / under	Wins on	Landed
36.5 / 38.5	37 or 38	5.7%
38.5 / 40.5	39 or 40	5.4%
40.5 / 42.5	41 or 42	5.3%
42.5 / 44.5	43 or 44	7.3%
44.5 / 46.5your two numbers	45 or 46	4.5%
46.5 / 48.5	47 or 48	6.0%
48.5 / 50.5	49 or 50	4.3%
50.5 / 52.5	51 or 52	5.9%
52.5 / 54.5	53 or 54	3.4%
54.5 / 56.5	55 or 56	3.3%

The mark on each bar is the 4.8% break-even rate.

A window covering totals of 43 and 44 has landed on 7.3% of 6,720 games; one covering 55 and 56 on 3.3%. Same width, same prices, and one of them clears the break-even rate by half again while the other does not reach it.

The likeliest single outcome of a middle attempt is the small loss, not the double win: at a 5% hit rate you are paying the miss 19 times out of 20. Prices also move while the second bet is going on, and a leg that does not get matched leaves a one-sided position rather than this table. Price the way out of an unmatched leg or take the margin out of both prices.

Written and reviewed by Alexander Korsager · last reviewed September 28, 2026 · corrections

The idea

A middle is two bets on the same market taken at two different numbers, so that a result landing in the gap between them wins both. If a total opens at 44.5 and moves to 46.5, betting the over at the old number and the under at the new one wins twice on 45 and 46, and wins one bet either side of that.

It is not an arbitrage and it is not a hedge.An arbitrage returns the same amount whichever way the market settles, so nothing about it can win twice. A hedge is a second bet placed against a position you already hold, to lock in or cap what it is worth. A middle is an opening play, and the price of the double win is that you lose the margin on both bets every time the number misses the gap.
That is the number this page leads with.The payout on the window is the exciting figure and the rarely occurring one; the honest question is how often the result has to land in the gap before the attempt pays for itself, and with the stakes balanced that answer is exactly the combined margin on the two prices. Two sides at -110 break even at 4.76%, whatever you stake.
Two results this tool prices that most middle calculators do not: a push, where a whole-number line lands exactly on the number and refunds that bet while the other one wins, and a wrong-way build, where the two numbers are entered the other way round and the gap loses both bets instead of winning both.
The break-even rate is only half of the comparison, though, and the other half is the one every middle calculator leaves to the reader: how often the gap actually lands. That number is not a matter of opinion for the sport most middles are struck in, so this page now reads it off 6,720 NFL regular-season games from 2000 through 2025 and prints it beside your own estimate. What it shows is that where the window sits matters more than how wide it is. Two sides at -110 two points apart break even at 4.76% whatever the numbers are, but a window covering totals of 43 and 44 has landed on 7.3% of games while 49 and 50 have landed on 4.3%: the same bet, at the same width and the same prices, on either side of break-even depending only on which two numbers the line moved through.

FAQ

Questions players ask

How often does a middle have to hit to be worth the attempt?

With the stakes balanced so both one-sided results cost the same, the break-even rate is the combined margin on the two prices. Two sides at -110 come to a 104.76% book, so the result has to land in the gap 4.76% of the time to break even, and the tool shows this figure for whatever prices you enter. Whether your gap clears it is a judgement about that specific number in that specific sport, which this tool does not make for you.

What happens if the result lands exactly on one of my numbers?

On a whole-number line that is a push: the bet on that number is refunded and the other bet wins, so you collect one winner and your stake back on the other. Half-point lines cannot push, which is why a 44.5 and 46.5 middle has three results and a 45 and 47 middle has five. The table on this page shows the push rows only when the numbers you entered can actually produce them.

Why are my two stakes different?

By default they are sized so both one-sided results cost the same amount, which is the same split the arbitrage calculator uses: each stake is proportional to its price's implied chance. It matters here because if the two misses cost different amounts, the break-even rate depends on which side the misses land, and nobody knows that split in advance. Set both stakes yourself and the tool shows a range rather than a single figure.

Can a middle lose both bets?

Yes, if the two numbers are entered the wrong way round: the over bet on the higher number and the under bet on the lower one. Then the gap between them is a window in which neither bet wins and the whole stake goes. It is an easy position to build by mistake and most calculators will price it without saying anything, so this one detects it and says so.

How often does an NFL middle actually land?

For a totals middle it is a matter of record rather than judgement, and the page prints it for whatever numbers you enter. Across 6,720 NFL regular-season games from 2000 through 2025, a two-point window on a total the market realistically hangs (roughly 36 to 56) has landed between 3.3% and 7.3% of the time depending on which two numbers it covers, against a 4.76% break-even at two -110 prices. Totals of 43 and 44 are the high end at 7.3%, and 55 and 56 the low end at 3.3%, so the same bet at the same width and the same prices can sit either comfortably above break-even or well below it. For a spread middle the same table gives only a ceiling, because the published frequencies count games decided by that margin either way and a spread middle needs one named side to win by it. Both readings are unconditional across all games rather than conditional on the number the market actually set. If you name the closing number and this site's own follow-up study reported a rate for it, the page swaps that unconditional reading for the measured one: a real difference for spreads, and a small one for totals.

Does a bigger gap mean a better middle?

A wider gap holds more results, so it lands more often, but the prices usually get worse as the line moves further, which raises the break-even rate. The two move against each other, which is why the frequency has to be judged against the break-even figure rather than in isolation. Note also that a one-point gap on whole-number lines, like over 45 and under 46, holds no result at all: 45 and 46 are pushes and nothing lands between them.

Methodology

How this is calculated

Each result is priced from the two bets directly: a bet wins and returns its stake times its decimal price, pushes and returns just the stake when the result lands exactly on a whole-number line, or loses the stake.

The break-even hit rate is the p that solves p x (both-win profit) + (1 - p) x (one-sided profit) = 0, which is L / (W + L) with W the profit when the gap lands and L the cost of a miss.

The historical figures beside your own estimate are counts, not a model: the tool takes every whole number strictly between your two lines, adds up the games in the sample that settled on those numbers, and divides by the sample.

An integer line is excluded from that sum because landing on it is a push rather than a win on both bets, which is the same window the branch table prices.

Naming the closing number the market actually hung at, where this site's own study reported a rate for it, replaces that unconditional count with the rate measured on games that closed at that exact number and width.

That measured rate is shown beside its own 95% Wilson interval, not as a bare percentage, since a reported cell is at most a few hundred games.

Common misreadings5

The both-win figure is not what the position is expected to return. It is the rare branch, and at a 5% hit rate the miss happens 19 times in 20.
A break-even rate being low does not mean the gap clears it. The two are independent, and putting a measured frequency beside the break-even rate makes the comparison possible rather than making it automatic.
A historical share is not a forecast, conditional or not. It says what has happened around a number across 2000 through 2025, not what will happen in one game, and adopting the figure as the estimate is a starting point rather than an answer.
The margin row is not the chance your spread middle lands, unless a closing number is entered and the conditional figure appears: the unconditional margin row counts both sides, so half of it or more belongs to the team you did not back, and reading it as your own number roughly doubles the edge you think you have.
Clearing the break-even rate on paper still leaves the variance: a middle is a long series of small losses interrupted by occasional large wins, which needs a bankroll that can sit through the losses.

Assumptions6

Results settle in whole units (points, goals, runs, games), which is what lets a whole-number line push and a half-point line never push.
Both bets are matched at the prices entered, for the stakes entered.
The default stake split equalizes the two one-sided results, which is what makes a single break-even rate well defined.
The historical frequencies are NFL regular-season games only, 6,720 of them from 2000 through 2025. Other sports and other leagues have their own shapes, and nothing on this page transfers a football frequency onto a basketball or a soccer market.
Those frequencies are unconditional, taken across every game in the sample rather than across games the market priced at these numbers. They are an anchor for the estimate the tool asks for, not a probability for one game.
The conditional figure shown when a closing number is entered comes from one closing consensus per game, not the two prices a middle actually needs. It is a rate on games that closed there, not a return on a bet built from two different prices.

Limits7

The frequency figures for numbers this site has not specifically studied are unconditional, and the direction that biases them is knowable. A game whose total the market hung at 44.5 lands on 45 or 46 more often than an average game does, so for a window sitting on the number the market actually set the unconditional share reads low; for a window well away from it, high. Where a closing number is entered and this site's own study reported a rate for it, the conditional figure above replaces that inference with a measurement, and how much conditioning actually moves the rate differs sharply by market: for a total it is small (the tightest total window measures 2.82% unconditional against 2.88% conditioned on its own closing number), for a spread it is not (the tightest spread window measures 4.02% unconditional against 4.80% conditioned).
The conditional dataset only reports a closing number when at least 100 games in the sample closed there: 24 of 78 distinct closing spreads and 27 of 64 closing totals clear that bar, covering 78% and 84% of games respectively. A number entered here that falls short of it is said plainly rather than interpolated from its neighbors or silently shown as the unconditional figure.
A reported cell's own conditional rate is still a point estimate off at most a few hundred games, and the 95% interval around it is wide: the Home +6, one-point-window cell (11 of 213 games) reads 5.16% with a 95% interval of roughly 2.9% to 9.0%, about six points wide (Wilson score interval, per #909's own re-derivation of this exact figure on the sibling research page). The page shows this interval beside the point estimate for whichever cell is reported, rather than letting the single percentage read as more settled than a sub-300-game count supports.
For a spread middle the published margin frequencies give a ceiling rather than an estimate, because they count games decided by that margin in either direction and a spread middle needs one named side to win by it. How that splits between favorite and underdog is not in the unconditional source; where a signed closing number is entered and reported, the conditional figure is already read against the named side and needs no split.
Football scoring is lumpy, so a frequency read off one window says little about the window a point away. That is a property worth using rather than smoothing over, and it is why the page prints the whole two-point range rather than one average.
The expected-profit figure ignores the push branches, because how often a total lands exactly on a whole number is a separate frequency the tool does not apply to your game. Pushes pay more than a miss, so on whole-number lines the figure understates the position rather than flattering it.
Execution is not modeled: prices move while the second leg is being placed, stakes get limited or voided, and an unmatched leg leaves a one-sided position rather than the table shown here.

Evidence

Every payout figure is a closed form derived from the two payout identities, with the derivations written out in math.ts. The branch table, both push cases, the wrong-way detection, the stake split against the arbitrage calculator's own split, and the worked -110 example are pinned in math.test.ts.

The historical frequencies are transcribed in nfl-history.ts from the frequency tables published at https://cleanuphitter.com/nfl/stats/nfl_common_scores.php, retrieved 2026-09-10, which give the number of NFL regular-season games from 2000 through 2025 settling on each winning margin and each combined total.

Only the integer game counts are transcribed, never the percentages, so both tables have to sum to exactly 6,720 games and a single mistyped digit fails a test rather than shifting a percentage nobody would notice; nfl-history.test.ts asserts both sums and then re-derives all 147 published percentages from the counts and checks each against the source.

The margin figures are also cross-checked against an independent compilation of the same seasons: BetMGM's own margin table, published 2025-09-30, gives 14.97% of games decided by exactly 3 points and 9.1% by exactly 7, against 14.99% and 9.09% here.

The conditional figures are read from our nfl-middles-2026 study's own figures.ts, generated from that study's own game-level run and re-derived from it by figures.test.ts, over 6,719 NFL regular-season games from the same 2000 through 2025 span sourced from nflverse/nfldata; reportedCellFor in math.ts matches a position's window width and an entered closing number against that dataset and is pinned in math.test.ts.

The 95% interval shown beside a reported cell's rate is the standard Wilson score interval, computed in reportedCellInterval (math.ts) from that same cell's own hits/games counts via the shared research-side wilsonInterval primitive (src/research/wilson-interval.ts), and is pinned in math.test.ts against the exact Home +6 cell (11/213) that #909 independently re-derived three times on the sibling research page.

A passing test means the transcription matches its source and the arithmetic matches the model, not that any of it forecasts one game.

Gambling can cause serious harm. The National Problem Gambling Helpline is free, confidential and open 24/7, call or text 1-800-MY-RESET (1-800-697-3738), or chat at ncpgambling.org/chat.

Read up

Go deeper

Related tools