The Lab · Our method · updated 2026-10-22

We tried seven ways to beat the market. Here's what happened.

Line shopping, sharp books, stale prices, unders, line moves, spreads and arbitrage — tested on 26,000 games with the rules written down first. Six didn't work, one only paid for itself, and one lead survived a later dig.

Most betting advice is a system. Bet unders. Follow the sharp money. Shop every line. Fade the public. Each one comes with a story about why it works, and almost none of them come with a test.

So we ran the tests. Not on a hunch or a hot week: on about 26,000 games of purchased odds across the NHL, NBA, NFL and MLB, 2021 through 2026, with every book's price at bet time and every final score.

Numbers as of October 10, 2026.

The rules we held ourselves to

A result only counts if it survives three things, and we wrote each rule down before running it:

1. Find it in one period, confirm it in another. Every idea was built on the 2021–23 seasons and then checked on 2024–25, which it had never seen. An idea that only works in the data it was found in is a description, not an edge.
2. Beat a price you can actually get. Every bet is scored at a real posted price — the best one on the board, and for most tests the best at a US-regulated sportsbook — never at a theoretical fair price.
3. Count the tries. Test enough ideas and one will look brilliant by luck. We counted every test and raised the bar to match (more on that in reading a null result).

The seven ideas, and how they did

1. Shopping for the best price

The most common advice in betting: find the book with the best number. We bet every side where one book's price beat the consensus by 2% or more. It looked great on the NBA in 2021–23 and then lost 13.9% on new data.

The reason has a name: adverse selection. When one book's price is out of line with everyone else's, it is usually the book that knows something — a scratched star, a goalie change — and the rest haven't caught up. The "best price" is often the best-informed price, and betting into it means betting against the news.

2. Following the sharpest book

If the consensus is the wrong truth, use the sharpest book instead. We ranked every book by how accurate its prices were on 2021–23 games, comparing each book with the others on the same games. LowVig came out on top. Then we bet US books whenever their price beat LowVig's. Up 3.0% on the period it was chosen on, down 8.5% on the next.

3. Catching stale books

A book whose price hasn't moved in twenty hours, while every other book moved, isn't better informed. It's asleep. That should be the cure for idea 1. It might be, but US books are rarely that slow: even after the vig, we found 13 usable cases in three seasons and 8 in two. That's too few to test, and too few to bet.

4. Betting the under

The story: casual bettors love overs, so books shade them, so unders pay. Totals turned out to be priced almost perfectly. Unders won 50.2% of the time against a priced 49.9%, in both periods. Taking the best price instead of the standard −110 cut the loss from 4.0% to 2.8%. It's still a loss.

5. Following the line move

This is the one real effect we found. When a line moves overnight, it tends to keep moving the same way, and the side it moved toward wins about 1.5 points more often than the morning price says. That showed up in both periods.

But at the best US price it returned +1.0%, then +0.2%. It roughly cancels the sportsbook's cut and no more. Useful as timing advice (if you're going to bet a side the line is moving toward, earlier beats later). Not a profit. Our own board found the same thing on a smaller scale: line movement leads the close and still doesn't pay, and adding it to a probability makes the probability worse.

6. Trusting the spread over the moneyline

The theory: sharp money lives in point spreads, so when the spread implies a different win probability than the moneyline, the moneyline is wrong. Betting that returned about +5% in all four cells, NBA and NFL, both periods. That's a promising pattern.

But each cell was within noise, and win-loss records take years to confirm a 5% edge. So we ran a faster test that uses price movement instead of results. If the spread really knows more, the moneyline should move toward it while the spread stays put. In 2021–23 it did. In 2024–25 both markets moved toward each other, which is just noise settling. The lead died there.

7. Arbitrage

Backing both sides at two different books for a guaranteed profit. Across about 52,000 market-games, the best US prices left an arbitrage 96 times — 0.2% — and the median one was worth half a cent on the dollar. Even taking the best price on both sides, the books kept 2.5% to 4.5%.

What survived

Nothing above made money at a price you can get. Later the same day we tested our own data on underdogs, and one thing held up:

In the NHL, the rested team against a team playing the second night of a back-to-back won 59.5% of 1,385 games when the price said 55.0%: +6% at the best US price, positive in all five seasons.

Even at the price five minutes before puck drop, it was still +5.0% across the last two seasons. The market moves toward the rested team during the day, but only by about half a point.

We don't trust it yet. It was found by looking, which is how flukes are found, and its most recent season was its weakest (+0.6%). Those are the warning signs of an edge the market is already closing. So we froze the rule on October 10, 2026, and we're scoring every game from that day on, at both prices, against its own record.

Status, October 22: still being scored forward, with the rule unchanged. Twelve days of hockey is far too few games to say anything; the result will be published here whichever way it lands.

What this means if you bet

• The price is the best forecast we've found. Every system above was a story until it was tested. Ask anyone selling one for a test like these: written down first, confirmed on data it never saw, scored at a real price.
• Shop for prices anyway. It won't make money on its own, but it cuts what each bet costs you. For example, shopping took the loss on unders from 4.0% to 2.8%.
• A hot week isn't evidence. We once watched a reading go four-for-four on a night's slate. Across 26,000 games it pointed the wrong way. See how many games until a record means anything.
• When the line moves your way, earlier is better. The overnight move tends to keep going.

Six of seven ideas not working isn't a disappointing result. It's the market doing its job, and it's why we publish the misses alongside the one lead we're still testing.

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