About — LockSZN

What this is

LockSZN posts model lines for NFL and college football — a spread and a total for every game we cover — before kickoff, then grades them after the final. Every play we actually post is graded at the number it went out at, never at a later one. Nothing is edited after posting and nothing is deleted: the Record page is the full ledger, wins and losses both, and last season's replay is in the archive.

How the model works

Two models per sport, trained on play-by-play back to 2015. The spread model predicts the home margin from our own numbers: per-play stats we compute from the raw play-by-play (success rate, yards per play, explosives, turnovers, field position), opponent-adjusted; our power rating from the rankings page; and a QB layer for who starts and how he has played. College adds last season, returning production and recruiting early in the year. No EPA, no outside ratings and no betting lines go in, so its disagreements with the market are real disagreements. The totals model predicts combined points from scoring form, pace, EPA, explosiveness, and drive finishing, plus venue facts (domes, and forecast wind for NFL). Win and over/under probabilities come from walk-forward calibration — every historical claim on this site was computed using only games finished at the time.

Cover probability is calibrated against the line you would bet into, not the one the game closes at. NFL: against Monday-regime opening lines (2018–25), where the record backs about 42% of our disagreement, so bet early; a play also needs a 2-point disagreement. College: against the close, where it backs about 16%; no call in weeks 1–4 or past 7 points. EV is that probability at the best DK or FanDuel price on our number.

Which number

Three numbers appear on the site and they are not interchangeable. A play is graded at the posted number it went out at, never a later one. The model book — every board number, play or not — is scored on its miss against the market at kickoff, with the market comparison taken at the early-week line, our first number after the week's lines open. How far the market then moves toward us by the last capture is line value, the model's leading grade.

The long version

First seen is the first DraftKings or FanDuel line we captured; last capture is the last pregame number, shown with how long before kickoff it was taken. We poll the market on a schedule, not continuously, so we never call the last capture “the close”. The early-week line is our first complete model run after the week's lines opened. Line value reads at fifty games where a win-loss record needs a season, which is why records against the spread ride behind it.

Board prices refresh at 7:05 AM ET daily, at noon and 6 PM on game days, and inside the hour before kickoff. Record: grades land within about 35 minutes of a final. Every page shows when its own data was written. Times shown are Eastern. None of this is betting advice.

Historical context

every season 2018–25 replayed against the reference close

The live record on /record grades the early-week line and the last capture; this replay grades a real closing line, so the two are not the same test. It updates after a reviewed backtest, separately from live results.

College

2018–2025 · walk-forward replay · graded at the closing line

College spreads stay Lean-only: this history has not established an edge. Graded at closing lines, not the early-week line the live record uses.

CFB spreads · average error in points · ATS excludes pushes
Week band	Games	Our MAE	Market MAE	ATS	Edge ≥3 pts
Week 1	336	13.22	12.49	52.3% 174–159–3	55.7% 118–94–2
Weeks 1–2	683	12.91	12.18	51.0% 341–328–14	53.1% 215–190–8
Week 4+	4,970	12.48	12.22	51.1% 2495–2384–91	51.6% 1031–966–38

Week 1336 games

ATS52.3%

174–159–3

Edge ≥355.7%

118–94–2

Our MAE13.2

Market MAE12.5

Weeks 1–2683 games

ATS51.0%

341–328–14

Edge ≥353.1%

215–190–8

Our MAE12.9

Market MAE12.2

Week 4+4,970 games

ATS51.1%

2495–2384–91

Edge ≥351.6%

1031–966–38

Our MAE12.5

Market MAE12.2

Records are wins–losses–pushes. Break-even at −110 is 52.4% of decided bets. Week 1 is also included in Weeks 1–2; these rows are not additive.

Method, uncertainty & backtest version

Each week was predicted using earlier completed games. Both MAEs use the same games with a prediction, final margin, and closing line. FCS matchups are excluded. Edge means the absolute difference between our predicted margin and the closing line. Zero disagreements produce no ATS pick. Model choices were informed by these historical seasons, so this is not an untouched validation set or a replay of posted bets. Positive line movement alone does not establish profitable betting.

Approximate 95% win-rate intervals for 3+ point edges (pushes excluded): Week 1: 48.9%–62.2%; Weeks 1–2: 48.2%–57.9%; Week 4+: 49.4%–53.8%. These intervals do not account for model selection or dependence between games.

Version week-bands-v2 · cfb_scratch · source preds_cfb_scratch.parquet · fingerprint 87c4af5a5b08. This snapshot updates after a reviewed backtest, separately from live results.

NFL

2018–2025 · walk-forward replay · graded at the closing line

NFL plays need positive EV at the price and a 2-point disagreement; a historical win rate alone qualifies nothing. Graded at closing lines, not the early-week line the live record uses.

NFL spreads · average error in points · ATS excludes pushes
Week band	Games	Our MAE	Market MAE	ATS	Edge ≥3 pts
Week 1	128	9.85	9.89	57.1% 72–54–2	57.1% 24–18–1
Weeks 1–2	256	9.20	9.21	55.6% 139–111–6	62.4% 53–32–2
Week 4+	1,843	10.08	9.87	52.1% 937–862–44	53.8% 250–215–11

Week 1128 games

ATS57.1%

72–54–2

Edge ≥357.1%

24–18–1

Our MAE9.8

Market MAE9.9

Weeks 1–2256 games

ATS55.6%

139–111–6

Edge ≥362.4%

53–32–2

Our MAE9.2

Market MAE9.2

Week 4+1,843 games

ATS52.1%

937–862–44

Edge ≥353.8%

250–215–11

Our MAE10.1

Market MAE9.9

Records are wins–losses–pushes. Break-even at −110 is 52.4% of decided bets. Week 1 is also included in Weeks 1–2; these rows are not additive.

Method, uncertainty & backtest version

Each week was predicted using earlier completed games. Both MAEs use the same games with a prediction, final margin, and closing line. Postseason games are included in Week 4+. Edge means the absolute difference between our predicted margin and the closing line. Zero disagreements produce no ATS pick. Model choices were informed by these historical seasons, so this is not an untouched validation set or a replay of posted bets. Positive line movement alone does not establish profitable betting.

Approximate 95% win-rate intervals for 3+ point edges (pushes excluded): Week 1: 42.2%–70.9%; Weeks 1–2: 51.7%–71.9%; Week 4+: 49.2%–58.3%. These intervals do not account for model selection or dependence between games.

Version week-bands-v2 · nfl_scratch · source preds_nfl_scratch.parquet · fingerprint 1500cda04442. This snapshot updates after a reviewed backtest, separately from live results.

Glossary

the words on the rankings, the cards and the record

Rating: College: our power rating: points better than an average FBS team on a neutral field. Built from how teams played (the scoreboard and the stat line), opponent-adjusted, with a preseason starting point from last season, returning production, talent and the portal. NFL: our power rating: points better than an average NFL team on a neutral field. Built from how teams played (the scoreboard and the stat line), opponent-adjusted; each team starts the season at 60% of last season's rating.
Per-play stats: EPA per play, success rate, pass and rush EPA, and explosive-play rate, for the offense and the defense. This season only, garbage time removed. The rankings table's Stats view ranks every team on each of them; Raw shows them as played, Adjusted takes the opponents out — college only, where it measurably beats the raw number at predicting the rest of the season.
Hidden points: the part of a margin the offense did not move the ball for: scores the offense did not make, safeties, the turnover column, kicking and fourth down over expectation, field position, and drive-extending penalties. Points per game against the opponent's. It barely repeats, so a team well above zero is a regression candidate. Credit: Carroll, Thorn & Palmer (1988).
Luck: actual wins minus earned wins. Earned wins — each played game's chance of being won given how it was played, summed.
Résumé: wins above average: wins minus the wins an average team would expect against the same opponents at the same sites. Results, not strength; margins don't count. Greyed until four games.
SOS: average Rating of opponents played. College: an FCS opponent counts as 15 points worse than the weakest FBS team. Greyed until four games.
Line value: How far the market moved toward our side by the last capture — from the first seen number, and from the early-week line. A signal, never a grade.
Early week: Our first number after the week's lines open: Sunday afternoon for college, Sunday night for the NFL. Line value is measured from here to the last capture.
First seen: The first DraftKings or FanDuel number we captured for this game.
Last capture: Our last pregame odds pull, usually inside the hour before kickoff. A median of the books we track, not a sharp close.
Posted number: The line and price a play went out at. The only number a public record is graded on.
Closing line: The reference closing line from CollegeFootballData and nflverse — a real close, used only for backtests and the 2025 replay.

Data & credits

the sources behind every number

nflverse — NFL schedules, play-by-play, results, depth charts, and injury reports. Used under CC BY 4.0.
CollegeFootballData.com — College schedules, play-by-play the archives miss, rosters, returning production, recruiting, the transfer portal, and the AP poll.
cfbfastR · sportsdataverse — College play-by-play archives (our totals model rebuilds EPA from them).
The Odds API — Betting line snapshots — the market numbers on the board and the movement charts.
Open-Meteo — Weather forecasts transformed into NFL totals-model inputs. Used under CC BY 4.0.
nfeloqb — Projected NFL starting QBs, so the model prices the QB who will play.
Over The Cap — NFL contract data for the offseason roster context.

Team logos are sourced from ESPN and CollegeFootballData; all team names and marks belong to their teams and schools. Historical opening-line research behind our methods also used Action Network's public scoreboard data (research only — nothing from it is served on this site).

Where a rule on the trends page was published by someone else, the credit travels with it — onto that trend's page and onto anything we post about it.

High totals go under — The rule was published for pro football by Rodney J. Paul and Andrew P. Weinbach. The idea is theirs; the college test and the live ledger are ours. Original paper.

Inter by Rasmus Andersson and Source Serif by Adobe are used under the SIL Open Font License 1.1.