The Lab · Reading the market · updated 2026-08-15

Why books disagree — and whether the softest line is the best bet

Sportsbooks differ because they run inventory, not forecasts. Here's what a soft price actually tells you, worked through one of ours that won and still failed the test.

Ten sportsbooks, one baseball game, ten different prices. If the market were a single wise forecaster, that shouldn't happen. The folklore explanation is that the odd one out knows something.

Sometimes. Mostly not. And either way, why they disagree isn't the question that decides anything. The question that decides things is whether a soft price is a reason to back that side.

It isn't. Here's the separation the whole subject hangs on.

Best price and best bet are two different claims

Best price. You have already decided to back the Athletics. One book pays +240, another +215. Take the +240. This is free. It requires no theory, it can't cost you anything, and it is the least controversial money in betting.

Best bet. The existence of +240 at one book is evidence that the Athletics are worth backing.

The first is arithmetic. The second is a prediction, and it needs support the first one never asked for. Collapsing them together is how a bettor ends up with a slate chosen entirely by whichever book happened to be slowest that morning.

Why books disagree at all

A sportsbook is not an independent forecaster competing to be accurate. It runs an inventory business — it holds positions, it wants a margin, and its posted number is a tool for managing what it's holding. Four ordinary reasons one game gets four prices:

1. Lopsided action. Money piles onto one side, so the book shades the price to make the other side more attractive. That shade describes its book, not its opinion.
2. Different risk appetite. A book with a $500 limit can sit on a number that a book taking $50,000 can't afford to.
3. Originators and followers. A few books make markets. Most copy them, with a lag and a house adjustment.
4. Plain latency. Nobody there has updated the number yet.

That last one matters most, and it's the one nobody wants to hear: a stale price is disagreement with no information in it. News lands, the field moves, one book hasn't gotten to it. On a screen that looks exactly like the informed kind — a longer number than everybody else. There is no visual difference between a book that's early and a book that's late.

One soft price, worked all the way through

Saturday, August 8, 2026. Caesars (it appears in our data as William Hill US) had the Athletics at +240. Everyone else in the field was shorter.

Start with what that price asks for. +240 means risking 100 to win 240, so break-even is 100 / 340 = 29.4%. When we flagged it, the de-vigged median of the field had the Athletics at 30.3% — about a point longer than the price demanded, which came to +3.1% by our arithmetic. That is a real outlier. It is the only kind of thing the board flags, and on the field's own number at that moment, it was the best price available by a clear margin.

Then the field moved, and it moved the wrong way. By the closing consensus the Athletics were 28.9% — 1.4 points below where they sat when we flagged the price. Two ways to state the miss:

• In probability points: the price demanded 29.4%; the close said 28.9%.
• In payout terms, which is how we grade it: a 28.9% chance is a fair price of about +246 (1 ÷ 0.289 = 3.46 in decimal odds), and the ticket paid 3.40. That's a shorter payout than the close said was fair. Graded at full precision the gap is −1.9% closing line value; recompute it from the rounded 28.9% above and you'll get −1.7%, which is the rounding, not a second finding.

If Caesars had been early and right, the rest of the field would have drifted out toward +240 by first pitch. It went the other way. Caesars wasn't ahead of the market; Caesars was off it — and the field spent the afternoon proving it.

A trap worth naming, because we nearly fell into it. Our board also lists Boston at 71.2% for this game — its most confident call of the day, which puts the Athletics at 28.8%. Put that next to the +240 and the pick looks like it never had an edge at all. It's the wrong comparison: the board shows the latest number, and 71.2% is the closing read, not the read when the price was flagged. Comparing a price you took at noon against a probability from first pitch will make any bet look mispriced in hindsight, in whichever direction the line happened to travel.

The Athletics won 7–3. We stake a fraction of Kelly rather than flat, so a +3.1% edge at this price sized to about 0.24 units, and 0.24 risked at +240 returns +0.58 units.

Read that pair of sentences carefully, because it's the most useful thing on the slate. The result went the way a bettor would have wanted, and the evidence still says the soft price stopped being value before the game started. A 29% shot comes in about three times in ten — that is what 29% means. One win doesn't confirm the flag, and a loss wouldn't have refuted it. The same board went 7–7 that day, and its single most confident call was the Red Sox side that lost.

The number that separates the two kinds of disagreement

You can't tell early from late by looking. You can tell afterwards, and cheaply, with closing line value.

• The soft book was early → the field converges on its number by close → CLV positive.
• The soft book was slow, or balancing its own action → the field moves off → CLV negative.

CLV is the number to watch because it's far less noisy than win/loss. It scores the price you got against a benchmark instead of inferring quality from one binary outcome. Our own published picks are 3 graded, 2–1, average CLV +3.66% — and three picks proves nothing whatsoever about profit. It's a number we can report honestly and cannot lean on.

Why this matters to us specifically

Here's the part that complicates our own product, so we'd rather say it than let a reader discover it.

We poll up to ten sportsbooks and won't call anything a consensus under three. With no signal promoted live, our published probability is the de-vigged median of the same books quoting the game. So when the board flags a price, it is not our opinion against the market's. It's arithmetic observing that one book pays more than the other books imply. That's a fact about the book, not a prediction about the game.

Which is why the panel on the board is now headed "prices out of line with the field" and says, in the panel itself, this is a price observation, not a prediction. It used to say "edge," and that word claimed a disagreement that did not exist.

It's also why three of those flags inside a single month were extreme underdogs. A flat EV floor needs less disagreement the longer the price gets — pure multiplication, worked out in expected value rather than repeated here.

We said we'd print this number. Here it is.

When this went up we owed you a measurement: how far apart the books actually get, across every game rather than the handful we flagged. We've since replayed 503 graded games at the moment the board looks at them each morning. The gap — best available price against the de-vigged median of the field, in probability points:

Two things fall out, and neither flatters the idea that shopping is a strategy.

The books essentially never disagree. Ten sportsbooks, 503 games, and the widest gulf between the best price and the field's own number was one and a third points of probability. Not one game in 503 reached two points. Whatever the folklore about finding the soft book, at this scale the soft book is barely soft.

The median is negative, which is the more useful half. On a typical game the best price available anywhere is still worse than what the field's own fair number says it should be. That's the vig, showing up as a measurement instead of an explanation. You are not shopping between a good price and a great one. You are shopping between a bad price and a slightly less bad one — which is exactly why it's worth doing, and exactly why it isn't a strategy.

It also settles something we were testing internally. We had built a second gate that would publish on gap alone, price-neutral, on the theory that it would stop our screen filling with longshots. It doesn't. At every threshold that returns any picks at all, the selection comes back 100% underdogs — the same bias, from a smaller sample. That gate is now switched off for good, and this is the paragraph explaining why.

What this still doesn't tell you

One game is an anecdote. August 8 shows the failure mode clearly; it does not tell you how often a soft price beats the close. The distribution above says how big gaps get, not how often the field converges toward one — those are different questions and we've only answered the first.

Sometimes the lagging book is the informed one. Nothing here says a soft price is always stale. It says you can't tell which you're holding at the moment you'd have to decide, and that the base rate favors "slow."

A one-point gap is inside our own noise. Across 456 graded games the market-derived favourite wins 55.7% (254–202), and by claimed confidence the buckets land close to where they should: 50–55% → 52.0% actual (n=204), 55–60% → 56.6% (n=152), 60–65% → 63.6% (n=66). Higher buckets are too thin to read. That calibration is the good news and also the caveat: the gap we flagged on August 8 was about 0.9 points, which the table above puts near the 99th percentile of every gap we've ever recorded — the fat end of a very thin distribution, not a different kind of event. A gap that size is worth noticing and is not worth much confidence.

Shopping does not rescue a losing strategy. Getting the best number lowers the price of everything you already do; it doesn't make a bad selection good. Our own line-movement test is the example: a real 0.46 points of value against a 1.5–2.25% fee still loses.

So the practical version is unglamorous. On sides you already wanted, take the longest price in the field — over a season that's real, compounding money, and it's the one form of value that doesn't require anyone to be right about a game. Just don't let a soft price pick your sides for you. When the field is tight and nothing clears, silence is the honest output.

Model output is informational and entertainment content, not betting or financial advice. If you bet, bet what you can afford to lose.

← Back to The Lab · How the model works · The record