The Lab
Behind every pick on this site sits an archive of odds snapshots — every game in the league, every legal US book, at two or three points before kickoff, since 2020. That's enough to answer questions beyond "did the pick win." Each study below states its method and its limits; all of it recomputes from the raw archive as new data arrives.
What is line shopping worth?
best legal-book price vs the market median at (near-)close · 8,415 eventsSportsbooks don't charge an entry fee — the commission is hidden inside the odds. It's called the vig (or "juice"), and it's why betting loses by default: win exactly half your bets at standard prices and your bankroll still shrinks. But books don't all hang the same number on the same game, which means there's an edge available before any question of picking winners: check several apps and take the best price. This study measures what that was worth.
Moneylines: extra payout from taking the best price
Standard vig runs ~4.5% — shopping claws back a third to three-quarters of the house edge before you've picked a single winner.
Who hangs the best moneyline price
Count of events where each book offered the single best price. DraftKings — the book most often quoted on air — is third.
Spreads & totals: the half-point on the shelf
| Market | sides | ≥½pt better line available | avg line edge | juice edge at same line |
|---|---|---|---|---|
| MLB spreads | 1,906 | 10% | 0.23 pt | +1.79% |
| MLB totals | 1,906 | 15% | 0.08 pt | +1.37% |
| NBA spreads | 3,284 | 47% | 0.27 pt | +1.07% |
| NBA totals | 3,284 | 65% | 0.44 pt | +0.83% |
| CBB spreads | 3,710 | 38% | 0.25 pt | +0.92% |
| CBB totals | 3,730 | 50% | 0.36 pt | +0.70% |
| CFB spreads | 3,832 | 51% | 0.43 pt | +0.94% |
| CFB totals | 3,794 | 57% | 0.53 pt | +0.77% |
| NFL spreads | 3,334 | 41% | 0.50 pt | +1.39% |
| NFL totals | 3,334 | 48% | 0.75 pt | +1.04% |
| NHL spreads | 104 | 5% | 0.13 pt | +2.25% |
| NHL totals | 104 | 31% | 0.17 pt | +1.34% |
| WNBA spreads | 178 | 48% | 0.28 pt | +1.15% |
| WNBA totals | 178 | 62% | 0.38 pt | +0.89% |
Method: each event's latest pre-kick snapshot (usually 0–10h out — near close, not the final tick); median across the legal US books carried by our source (offshore excluded); a side needs 3+ books quoting it. The median baseline understates shopping value vs the realistic alternative of one fixed app. Sample skews toward days the shows had picks. Generated 2026-07-08.
Does it matter which book we measure against?
the same 7,150 picks, graded three times with three different rulersThe wrinkle: "where the line closed" depends on which book you ask. Every closing-line number on this site is measured against the consensus close — the median across legal US books. That's a choice, and it invites a fair objection: nobody can actually bet the median, and books disagree with each other at close all the time (see the shopping study above). So maybe the "sharpness" only exists against our invented yardstick. To test that, we re-graded every pick against two real books' closing lines instead — same games, same snapshots, same moments. If our numbers were an artifact of the yardstick, these three beat-the-close rates would disagree:
How often each personality beat the close, under each ruler
| Personality | vs consensus | vs DraftKings | vs FanDuel | biggest shift |
|---|---|---|---|---|
| Brandon Anderson | 71.9% n=452 | 72.4% | 73.2% | 1.3pp |
| Stuckey | 71.2% n=548 | 68.6% | 68.5% | 2.7pp |
| Chad Millman | 65.2% n=836 | 64.9% | 65% | 0.3pp |
| Chris Raybon | 64.9% n=425 | 65% | 65.8% | 0.9pp |
| Simon Hunter | 61.6% n=1520 | 60.9% | 60.6% | 1.0pp |
| Chris Canty | 49.8% n=223 | 53.3% | 49.8% | 3.5pp |
| Nick Wright | 49.2% n=250 | 51% | 50% | 1.8pp |
| Joe Fortenbaugh | 48.2% n=110 | 48% | 48.4% | 0.2pp |
They don't disagree. Site-wide, the three rulers land within half a point of each other, no personality's rate moves more than a few points ("biggest shift" below), and nobody changes places — whoever is sharp against the median is sharp against DraftKings and against FanDuel. Book-level disagreements are noise around the market's center, and across thousands of picks they cancel out.
Method: identical picks, snapshots, and timestamps in all three columns; when a book simply wasn't quoting a game in the snapshot, the pick drops out of that book's column rather than substituting another price — which is why the sample sizes differ slightly. Table shows personalities with 100+ decided moves. Generated 2026-07-08.
Do shows pick games that were already moving?
line movement on picked games vs unpicked games from the same slatesAverage spread movement between first and last pre-kick snapshot
In the NBA, CBB, and MLB, picked games moved the same or less than their unpicked slate-mates — the beat-the-close rates aren't an artifact of gravitating to volatile games. The exception is college football, where picked games moved ~50% more; CFB beat-close claims deserve that asterisk.
Method: consensus (median) home spread, total, and moneyline compared between each event's earliest and latest pre-kick snapshots at least 2h apart; "picked" = any show's captured pick joins to the game. Movement between two coarse snapshots understates total intraday variation. The NFL comparison is weak — the shows pick so much of each NFL slate that only 63 games were left unpicked, which is itself a finding. Generated 2026-07-08.
All studies recompute from the raw snapshot archive with npm run studies — zero API calls.
Related: steam moves · CLV methodology.