SharpEdge

The Ledger

We went 2-3 this week. Both wins came on bets where we got a better number than the market closed at. All three losses came on bets where we got exactly the market's number. That split matters more than the record, and it's what this issue is about.

Every play was one unit at DraftKings at -110, tagged either Play (our stronger call) or Lean (the lighter one).

The spread: The head start the underdog gets. Seahawks -3 means Seattle has to win by more than 3 for that bet to cash. The total: The over/under on combined points from both teams.

Closing line value is CLV: whether the line you bet was better than where the market closed. The best long-term sign a bettor knows what they are doing, because closing lines are the sharpest numbers there are. Units: A fixed bet size so results can be compared. One unit is one unit whether that is $10 or $100. Hit rate: The percentage of bets that won. At standard -110 odds you need 52.4% just to break even.

Break-even is 52.4%. We're at 20% and down 6.18 units on 10 graded plays. That's a bad start and I'm not going to dress it up. The only number on that row that isn't ugly is the +0.4 average CLV: on average, our number beat the close by 0.4 points. That number gets the same discount as the record, because both are subject to regression: Pulling an extreme number back toward average because small samples lie. A 6-0 team is good; it is not undefeated-good. Ten bets don't prove we're bad, and they don't prove we're sharp.

What the Closing Line Said

Two plays beat the close: IND +3, which closed +1.5, and DET -6.5, which closed -7. NE +3 and both unders closed exactly where we bet them. None of our plays closed worse than our number.

That line matters more than the scoreboard. One game can turn on a tipped pass, and five bets, or ten, aren't nearly enough to wash that out. The closing line is the market's final answer after every sharp bettor and every model has had a say, and it's the best estimate anyone has of what a game is worth. If you get a better price than that final answer week after week, you're buying at a discount, and over a long run the discount turns into money. If you pay the market's price, you're paying full retail plus the book's cut, and no read on a matchup outruns that for long. The scoreboard grades luck and skill mixed together. CLV grades the part we control.

There are two more things I'd rather say out loud than bury. First, both Plays closed at zero CLV and lost, while the Leans went 2-1. If the Play tag keeps landing on bets the market never moves toward, the tag isn't telling us anything.

Second, our backtest replays the model on 2023 through 2025 using only what was known at the time. Betting only when the model disagreed with the closing spread by 2 or more points, it went 51.7% against the spread over 437 plays. That's under break-even, so the model alone isn't the edge. Key numbers are: NFL games land on 3 and 7 far more than any other margin, so a half point on either side of those numbers is worth real money. Both winners were key-number bets: +3 before the market went to +1.5, and -6.5 before it went to 7. NE +3 sat on a key number too, and it closed right there. Being on 3 isn't the same as getting the good side of 3.

The Lean, Revisited

The pack doesn't include final scores, so I won't tell you how close any of these games were.

NE +3 at JAX (Play, loss): the read. The number was fair, since it closed at 3. The problem was the rating behind the bet. A power rating is A number for how good a team is, in points, against an average team on a neutral field. Subtract two ratings, add home field, and you have a model spread. Ours has New England at 3.3, fifth in the league, and our ratings blend 30% this season with 70% last season. Last season the Patriots went 12-5 against the spread and beat it by 6.2 points a game. This season they're 1-1-1.

The current profile has the Patriots' offense at -0.146 EPA per play. EPA per play is Expected Points Added: how many points a play moved the team toward scoring compared with an average play from the same spot. Positive is good on offense, negative is good on defense. That offense ranks 28th, turns it over 2.67 times a game and scores 12.0 points a game. Jacksonville ranks second in net EPA per play, offense and defense combined, and allows 11.7 points a game. Those numbers include this game, so some of it is hindsight. The miss is still clear: we trusted last year's Patriots, and the market never moved to agree with us.

Under 42.5, CIN @ PIT (Play, loss): probably variance, and I can't prove it. It closed at our 42.5. The read still looks sound: Pittsburgh's offense is 30th in EPA per play at -0.169, and Cincinnati's defense is fifth. The risk was always Cincinnati's offense, which scores 26.7 points a game. But zero CLV means the market thought 42.5 was exactly right. A bet at the market's price is close to a coin flip before the book's cut. Losing a coin flip isn't a sin. Tagging one as a Play is the part I own.

Under 43.5, ATL @ GB (Lean, loss): the read. It was a fair number that closed at 43.5. The miss was Green Bay's defense. The current profile has it allowing 30.3 points a game, 22nd in defensive EPA, and giving up +0.146 EPA on pass plays, which is bad for a defense. Those numbers include this game, so cause and effect are tangled. I'd still rather own the miss than argue it.

IND +3 vs HOU (Lean, win): read and number. The market moved 1.5 points our way. This is the bet I want every week, though the cover itself still took some luck.

DET -6.5 vs NYJ (Lean, win): the number. Detroit is 0-2-1 against the spread this season in the pack, so by the line those records use, Detroit did not cover this game. We cashed because we laid 6.5 before the line went to 7.

Looking Ahead

These are the games where DraftKings opened on 3 or 7, or a half point away. "Model" is our model spread: the rating gap plus 1.8 points of home field. "Edge" is how far the model sits from DraftKings, and which team it favors.

The backtest tells us how much to trust each edge. When the model disagreed by 2 to 3 points, it hit 54.7%. From 3 to 4.5 points, it hit 52.1%. At 4.5 or more, it hit 47.7%, which means the model was wrong more often than right.

Worth a look, in the 2-3 point band:

• HOU -2.5 vs DAL. The model has Houston by 4.7, and laying 2.5 keeps you under 3. Dallas is 32nd in defensive EPA and gives up +0.408 EPA on pass plays. What would make me wrong: Houston's offense is 24th, and the Texans are 0-3 against the spread. At -3, I pass.
• WAS +3.5 vs IND, at Tottenham Hotspur Stadium. It opened at 3 and moved to 3.5, so Washington backers now get the good side of 3. Indianapolis is 31st in defensive EPA. The worry is that Washington allows 29.7 points a game itself.
• CAR +3.5 vs DET. This one is already on the good side of 3. Detroit's defense is 29th and allows 31.7 points a game.
• CLE +2.5 vs PIT. The model calls it close to even, but 2.5 is the wrong side of 3 for a Browns backer. It's a division game, with both teams on 4 days of rest.

Big disagreements, where history says the model is usually wrong: JAX, CHI (and CHI -3 is already gone), TB and NE. The NE edge comes from the same Patriots rating that just cost us. I'm not chasing it.

On the number with no argument: DEN @ SF, LAC @ SEA and ATL @ NO. The model is within 1.2 points of DraftKings on all three.

The pack only has DraftKings lines this week, so I can't tell you where a better half point lives. Line shopping: Checking the same bet at several sportsbooks and taking the best number. Getting +3.5 instead of +3 is free equity, and it costs nothing but a minute. On the Browns, that minute is the difference between passing and taking a look.

If I had to name one game to watch, it's HOU -2.5. The disagreement sits in the backtest's best band, and the number sits on the right side of 3. None of these are ledger plays yet.

Ledger Entries

Plain English

• spread: The head start the underdog gets. Seahawks -3 means Seattle has to win by more than 3 for that bet to cash.
• total: The over/under on combined points from both teams.
• closing line value: CLV: whether the line you bet was better than where the market closed. The best long-term sign a bettor knows what they are doing, because closing lines are the sharpest numbers there are.
• units: A fixed bet size so results can be compared. One unit is one unit whether that is $10 or $100.
• hit rate: The percentage of bets that won. At standard -110 odds you need 52.4% just to break even.
• regression: Pulling an extreme number back toward average because small samples lie. A 6-0 team is good; it is not undefeated-good.
• key numbers: NFL games land on 3 and 7 far more than any other margin, so a half point on either side of those numbers is worth real money.
• power rating: A number for how good a team is, in points, against an average team on a neutral field. Subtract two ratings, add home field, and you have a model spread.
• EPA per play: Expected Points Added: how many points a play moved the team toward scoring compared with an average play from the same spot. Positive is good on offense, negative is good on defense.
• line shopping: Checking the same bet at several sportsbooks and taking the best number. Getting +3.5 instead of +3 is free equity, and it costs nothing but a minute.

Sources

DraftKings lines via ESPN, including opening numbers from Sept 28. nflverse play-by-play and schedule data for team profiles, rest days, stadiums and projected starting quarterbacks. Official injury reports, which list no entries for these games. SharpEdge power ratings, model and 2023-2025 walk-forward backtest. SharpEdge graded ledger.

21+. If you or someone you know has a gambling problem, call 1-800-GAMBLER.