About

What the numbers mean

33 terms in 4 groups.

Our numbers

Model record: The model's own line, set before kickoff, against the closing spread, on every game it had a line for.; Not the Calls record. It counts every game the model had a number for, not only the games we published as calls, so it is the fairer test of the model.; Each game is graded once: if the model had the home team winning by more than the closing spread, it takes the home side, otherwise the away side. A game where the model's line equals the close has no side and is left out; a final that lands exactly on the close is a push.
Calls record: Only the games published as McConnell's Calls, graded at the posted price, flat stake.; Not the Model record. The calls are a handful of games a week, picked by rules written down in advance, so the two records count different games and can disagree.; Every call is frozen before kickoff and never revised. The NFL sheet stakes one unit a call; the college sheet stakes 0.25 units.
Rating (NFL): Team Elo plus the starting quarterback's adjustment. The number the game model predicts with.; Not a preseason power ranking. It moves with this season's results every week, and it is the one NFL team rating and rank the site shows.
Disagreement: How unusual the gap between our number and the market's is, compared with every other call we've made.; Not a win probability. A disagreement in the 92nd percentile says we have rarely been this far from the market. It says nothing about how often the pick wins.; We rank the gap between our number and the market's against five seasons of our own past calls. The 92nd percentile means only 8% of our calls have been further from the market than this one. A big disagreement can still be wrong. This measures distance, nothing else. For whether calls like this win, read the slot's record.
Points edge: How far our line sits from the market's, in points.; Not expected profit. Our model and the book see the game 14 points apart, and at least one of us is wrong.
Record score: How much our past accuracy on these two teams should make you trust this week's number, on a 0 to 5 scale.; Not a rating of the teams. It rates us, not them.; Built from how well we've predicted these specific teams in seasons already played, and how we've done on games this far from the market. A low score means our history here gives you no reason to take this week's call on trust. The call can still be right. We just haven't earned it.
Typical miss: How far off our predicted margin usually lands for this team, in points.; Football is noisy: even a good model misses by double digits routinely. If we say a team wins by 7 and we typically miss by 11, treat the 7 as a rough middle rather than a prediction.
Line runs: Which way we've been wrong on this team: plus means we rated them too highly, minus means too low.; Not the same as typical miss. Typical miss is how far we're off; this is which direction we lean.
Plausible range: Where the true number could reasonably sit, given how few games we're working from.; A record of 79-54 gives 59%, but 133 games isn't many. The range is what the real rate could plausibly be if we ran the same seasons again. When the range is wide, treat the headline number as a rough read.
Total range: The band of final scores we think a game is likely to land in, with the market's posted total marked on the same line.; Not a pick, and not a lean. We are not saying the game goes over or under. The band is the whole statement, and the gap between our middle and the market's number is not something we act on or publish.; The dark bar is where we think half of games like this finish; the lighter bar, four in five. The tick is the middle of our range, and it never appears without the band around it. Judge it on coverage: over a season, about four finals in five should land inside the 80% band. One landing outside is the band working as advertised, not a miss. The bands are wide, around 34 points on an NFL game, and that width is the honest finding, not a hedge. Game totals are roughly half as predictable as margins, for everyone.

The calls

Unit: One unit means one standard bet, whatever your normal size is.; Not a dollar amount, and not advice about how much to bet. We use units so the record means the same thing to everyone regardless of stake.
Hit rate: How often calls in this slot have won, in seasons already played.; Not a forecast. We found these slots by looking for what had already worked, so the rate describes games we have played, not the one coming up.; Two things to hold in mind. First, it's measured per slot, not per call. Every lean shares the same slot record. Second, the samples are small enough that the true rate could sit several points either side of the number shown, often close to break-even.
Posted vs latest: The number we published, next to the number available now. We grade ourselves at the one we published.; If the market moves toward a call after we post it, we were early to something the market later agreed with. If it moves off, a bettor today gets a better number than we did. We never re-price a published call, because grading yourself at a number you didn't take is how records get faked.
Value dog: The one underdog moneyline a week where our win probability sits furthest above the market's, after the vig is removed.; Not the biggest disagreement on the board. Only games inside the lean band qualify, and the price has to sit between 15% and 85% with the vig removed.; Ordered on the largest gap between our de-vigged probability and the market's, because that is the rule the replay of past seasons was measured on. One a week, flat one paper unit, graded at the price posted.
Lean: A spread call in the band where our number sits 1.5 to 3 points from the frozen line, taking our side of it.; Not a strong opinion. A lean is by definition a small disagreement; bigger ones are outliers, which we do not call.; Up to three a week. Games through a key number come first, then the closest to the market. Every lean is one flat paper unit, graded at the frozen line, never a later one.
Key-number call: One game a week where our number and the market's sit on opposite sides of 3 or 7, the two margins NFL games land on most.; Not a proven edge. It is labeled experimental because it is a forward test, written down before week 2 and running through week 18.; On seasons already played, leans that straddled 3 or 7 hit 57.5% against 51.0% for the rest, on a sample too small to clear the juice with confidence. The slot exists so the answer comes from games played after the hypothesis was written. Nothing about it can be tuned during the test.
Agree, lean, outlier: How far our line sits from the market's: under 1.5 points is agree, 1.5 to 3 is lean, 3 or more is outlier.; Not a confidence scale. An outlier is a game where our own error measurably exceeds the market's, so a bigger disagreement is a worse call, not a better one.; Only lean-band games reach the calls sheet on their own; agree-band games arrive only through the key-number slot, and outliers never do. Every football game page shows its band.
CLV: Closing-line value: the number we posted compared with the number the market closed at, in points for spreads and in probability points for moneylines.; Not the same as winning. A call can win with negative CLV and lose with positive CLV; over many calls, CLV says whether we are ahead of the market or behind it.; Positive means the market moved toward our side after we posted. Beating the close consistently is harder to fake than a win-loss record, which is why it is one of the two lines the kill bar reads.
Wins above market expectation: Wins minus the number the market's own prices said we should have, so zero is what no edge looks like.; Not a win-loss record. A sheet of heavy favorites posts a handsome record and loses money; a sheet of long dogs posts an ugly one and may not.; Each call logs the market's de-vigged probability of the side we took. Add those up and you get the wins a no-edge sheet would expect; the headline is the actual wins minus that. A standard error comes with it, and after 100 graded calls a result below minus 1.96 of those retires or reframes the sheet.
The Lock (retired on the NFL sheet): A favorite the model agreed with. Retired from the NFL sheet 2026-09-14; the college sheet still publishes one.; Not a guarantee, despite the name.; On the 2021 to 2025 replay it hit 75.6% and lost 9.8% on the money: a win-loss line that measured the market's favorite rather than the model. Week 1 of 2026 was graded under it and stays on the ledger as published.
The Dog (college sheet): The underdog moneyline closest to the market on the college sheet.; The NFL sheet replaced it with the value dog on 2026-09-14. Week 1 of 2026 was graded under the old slot and stays on the ledger as published.
The Total (college sheet): The college game where our calibrated total sits closest to the market's, called over or under.; Retired from the NFL sheet 2026-09-14, because the NFL game card publishes no total of its own and a sheet calling one contradicted the card. The college model behind this call has not yet been graded against closing lines; read it as untested.
The Board (college sheet): Spread calls filling out the college slate, closest to the market first.; Retired from the NFL sheet 2026-09-14: selecting the smallest disagreement inside the callable band is laying -110 to bet the market's own number. The lean slots replaced it there.
The Pocket (retired): A retired college slot: near-even games where our line sat ten or more points from the book's.; Kept here so archived weeks still read. Games that far from the market are now outliers, which no sheet calls.
The Home Dog (retired): A retired college slot: a home underdog getting only a small number of points.

How the models work

Season simulation: Plays the whole schedule thousands of times to get a team's win total, playoff odds and title odds.; Not a second opinion on one game. It answers season-shaped questions: how many wins, how likely to make the playoffs. Single-game numbers come from the game model instead.
Elo: A rating that moves up when a team wins and down when it loses, by more when the result was a surprise.; Borrowed from chess. Several of our ratings are built on it because it handles strength of schedule by itself: beating a good team moves you more than beating a bad one.
EPA: Expected points added: how much a single play improved a team's scoring chances.; A five-yard gain on 3rd-and-2 is worth far more than the same gain on 3rd-and-20. EPA prices that difference, which is why it describes offenses better than raw yardage.
Calibration: Whether things we call 70% actually happen about 70% of the time.; A model can pick winners well and still be badly calibrated, which means confidently wrong. We publish ours because it is the number hardest to fake.

Betting basics

Spread: The margin the favorite has to win by for a bet on them to pay out.
Total: The combined points both teams are expected to score.
Moneyline: A bet on who wins outright, with no margin involved.
Closing line: The last number available before kickoff.; It is the market's final answer. Beating it consistently says more about skill than any single week's win-loss record.