Baseline Sports

NFL betting analytics · every pick graded in public

A betting model
you can audit.

Baseline Sports is a statistical NFL research engine. Our published picks went 46–25 over the last two seasons — and instead of asking you to take that on faith, we built a platform that logs every model call, grades every result in public, and leaves the losses on the page.

Join the waitlistSee the method
Published picks, 2024–25 seasons
46–25
Published win rate · breakeven 52.4%
64.8%
Model bets vs market calls analyzed
102 / 1,139
NFL games in the dataset
2,493

46–25 is our published-picks record across the 2024 and 2025 NFL seasons (95% interval 53.2%–74.9%). Every model number on this page is computed from the platform’s graded ledger — out-of-sample backtests at closing lines for the two models shipped so far, with more of the stack still being ported in. Past results never guarantee future performance.

01

Losses stay published

Our models went 21–22 in 2024. That season is on this page in full — a track record you can filter is marketing, not a record.

02

Backtest labeled as backtest

The 2024–25 record is out-of-sample at closing lines, and says so. Live picks lock before kickoff, timestamped.

03

Intervals, not hot streaks

Even 102 model bets is a small sample. We print the confidence interval instead of pretending a hot rate is destiny.

04

We don’t take bets

Baseline Sports is research, not a sportsbook. We never hold your money and never profit from a losing bet.

01The method

One question, asked 2,493 times.

For every game, the models ask: when the market priced a matchup like this before, what actually happened? They ask it twice — once against the spread, once against the total.

  1. 1

    Project the margin

    Score projections estimate a final margin for both teams — built before the model ever looks at the betting line.

  2. 2

    Place it against the market

    Each game becomes one point: our predicted margin against the Vegas closing spread. History fills in the rest of the plane — 1,924 reference games from 2017–2023.

  3. 3

    Bet only lopsided history

    The model collects every historical game inside a tight window around today’s point. If those comparables covered far more often than not — or far less — that’s a pick. Otherwise it passes. It usually passes.

Model predicted margin versus Vegas closing spread, 1,924 reference games 2017 to 2023Scatter plot of every reference game. Blue dots are games where the home team covered, gray dots where it did not. A dashed ellipse marks the comparable-game window around one example game.-18-12-60+6+12+18-20-100+10One game’s comparable window223 historical comps insideModel predicted home margin (points)Vegas closing spread (home)
Real data, not an illustration: all 1,924 reference games in the spread model’s decision space (the totals model works the same way on its own axes). The dashed window shows the actual comparables for NYG @ WSH, Week 2, 2024 — 223 games: 91 covered, 123 didn’t, 9 pushed (a 0.69 cover ratio), so the model bet NYG +1.5. It lost. That loss is in the ledger below.

1,139 market calls analyzed102 bets

A spread call and a totals call for every 2024–25 game, and the models’ most common output is still “pass.” Edges against the closing line are rare — a model that bets every slate is a content machine, not a model.

02Track record

The whole ledger. Both seasons.

Most pick services show you their hot streak. Our published picks went 46–25 over two seasons — and below it is everything the two in-platform models did across 2024–25, drawdown included.

46–25Published picks, 2024–25 · 64.8%
60–42In-platform model bets · 58.8%
+12.55uModel net units · +12.3% ROI at −110
21–22The models’ 2024 season — shown anyway
Cumulative net units across all 102 graded model bets — spread and totals, in order. Flat 1 unit per pick at −110; drawdowns included, because they happened.
Cumulative net units across all 102 graded model bets, 2024 to 2025Line chart starting at zero, dipping to -4.1 units during the 2024 season, and finishing at plus 12.55 units after the 2025 season.-5u-1u0+3u+7u+11u2024 season2025 season-4.1u+12.55u
Complete model ledger · 102 bets (31 spread, 71 totals), 2024–25
SeasonWkGameResultPickFinal
202519HOU @ PITLOver 39.530–6
202518BAL @ PITWOver 4124–26
202518SEA @ SFWSEA −0.513–3
202518WSH @ PHIWOver 3924–17
202517NE @ NYJWOver 4342–10
202517NYG @ LVWOver 41.534–10
202517NYG @ LVWNYG +1.534–10
202517SEA @ CARWSEA −7.527–10
202516BUF @ CLEWOver 4223–20
202516CIN @ MIALUnder 50.545–21
202516MIN @ NYGLOver 4416–13
202515ARI @ HOUWOver 42.520–40
202515CAR @ NOWUnder 40.517–20
202515CLE @ CHIWCHI −7.53–31
202515LV @ PHILOver 38.50–31
202515LV @ PHIWPHI −12.00–31
202515NYJ @ JAXWOver 41.520–48
202515TEN @ SFWOver 44.524–37
202515WSH @ NYGLUnder 4729–21
202514PIT @ BALWOver 43.527–22
202514SEA @ ATLWOver 44.537–9
202513ATL @ NYJWOver 39.524–27
202513HOU @ INDWUnder 44.520–16
202513JAX @ TENLOver 4225–3
202512ATL @ NOLOver 39.524–10
202512ATL @ NOWATL +1.524–10
202512PIT @ CHILUnder 45.528–31
202511NYJ @ NELOver 4314–27
202510CLE @ NYJWOver 3920–27
202510LV @ DENLOver 42.57–10
202510NYG @ CHILNYG +3.520–24
20259DEN @ HOULOver 39.518–15
20259NO @ LARWLAR −14.010–34
20258GB @ PITLUnder 4535–25
20258NYG @ PHIWOver 4420–38
20258TB @ NOWUnder 4723–3
20258TEN @ INDLTEN +14.014–38
20257IND @ LACWIND +1.538–24
20257NO @ CHILNO +4.514–26
20256LAC @ MIAWOver 43.529–27
20256NE @ NOLNO +3.525–19
20256PHI @ NYGWOver 4117–34
20255DEN @ PHIWDEN +4.521–17
20255DET @ CINLCIN +9.537–24
20255SF @ LARWSF +8.026–23
20255TB @ SEAWOver 44.538–35
20255TEN @ ARIWOver 41.522–21
20254CAR @ NEWOver 43.513–42
20254CIN @ DENLOver 43.53–28
20254JAX @ SFWJAX +3.526–21
20253ARI @ SFWUnder 45.515–16
20253CIN @ MINWOver 42.510–48
20253LAR @ PHIWOver 44.526–33
20253LV @ WSHWOver 43.524–41
20252NE @ MIAWOver 43.533–27
20252NYG @ DALWNYG +4.537–40
20252SEA @ PITWOver 39.531–17
20252TB @ HOULOver 4320–19
20251KC @ LACLUnder 45.521–27
202421WSH @ PHILWSH +6.023–55
202419PIT @ BALLOver 43.514–28
202418NO @ TBWNO +14.519–27
202418NYG @ PHILNYG +3.013–20
202418SEA @ LARWOver 3830–25
202418SF @ ARIWOver 4324–47
202417IND @ NYGWOver 4033–45
202416HOU @ KCLHOU +4.019–27
202416TEN @ INDWOver 42.530–38
202415CHI @ MINLOver 43.512–30
202415PIT @ PHILOver 43.513–27
202414ATL @ MINLUnder 4621–42
202414CAR @ PHILPHI −13.516–22
202414GB @ DETLUnder 51.531–34
202414LAC @ KCWLAC +4.517–19
202414SEA @ ARIWOver 44.530–18
202413LV @ KCLOver 42.517–19
202413NYG @ DALLNYG +5.020–27
202412BAL @ LACLUnder 50.530–23
202412DAL @ WSHWOver 44.534–26
202412SF @ GBWOver 44.510–38
202411CLE @ NOWOver 4414–35
202411LV @ MIAWOver 43.519–34
202411MIN @ TENLOver 4023–13
202411WSH @ PHILWSH +4.518–26
202410MIN @ JAXWJAX +7.012–7
202410NYG @ CARWCAR +6.517–20
202410TEN @ LACWOver 39.517–27
20248ARI @ MIALUnder 46.528–27
20248BUF @ SEAWUnder 45.531–10
20247DET @ MINWDET +1.531–29
20247LV @ LARLOver 43.515–20
20247MIA @ INDLOver 4410–16
20247TEN @ BUFWOver 4110–34
20246IND @ TENWIND +2.520–17
20246TB @ NOWOver 42.551–27
20245CLE @ WSHWOver 4413–34
20245NO @ KCLOver 43.513–26
20244DEN @ NYJLOver 39.510–9
20243HOU @ MINWUnder 46.57–34
20243JAX @ BUFLJAX +4.510–47
20242NYG @ WSHLNYG +1.518–21
20241ARI @ BUFLUnder 4628–34

Finals shown away–home. Spread picks show the team and line; totals picks read Over/Under. Graded at standard −110; playoff weeks included (2024 Wk 21 is the NFC Championship). Members can export the full decision log — every analyzed market call including passes — as CSV.

03The product

What members get.

The weekly card

Every qualifying pick with its evidence attached: the comparable-game window, the cover counts, the ratio. When nothing qualifies, we tell you that too.

The full decision log

Every analyzed game — including the passes — exportable as CSV. Audit the model whenever you like; that’s what the export is for.

The live track record

The dashboard version of this page: every pick graded automatically as games go final, with streaks, intervals, and ROI — never edited after the fact.

04FAQ

Fair questions, straight answers.

Why should I trust a betting model on the internet?

You shouldn’t — not on our word alone. So we publish the material you’d need to check: the complete graded ledger with losses left in, the methodology in plain language, confidence intervals instead of streak-math, and a CSV export of every decision the model made, passes included. We also don’t take bets, so unlike a sportsbook, nothing about our business improves when you lose.

What’s the difference between the 46–25 and the 60–42?

46–25 is our published-picks record: what we actually posted for members across the 2024 and 2025 seasons, using our full stack — including models we haven’t ported into the platform yet. 60–42 is the fully auditable layer: the two models shipped in the platform so far (spread and totals), backtested out-of-sample against closing lines their reference history (2017–2023) never saw. We keep the two labeled separately because the distinction matters. Going forward, live picks are timestamped and locked before kickoff, then graded automatically when games go final.

Why do the models pass on 91% of their calls?

Because the closing line is hard to beat, and pretending otherwise is how betting content usually works. Across 2024–25 the models analyzed 1,139 market calls — a spread call and a totals call for every game — and found 102 spots where the comparable history was lopsided enough to act on. If you want a pick for every slate, plenty of services will sell you one — that’s entertainment, not analysis.

Do you guarantee I’ll make money?

No, and we’d encourage you to close any tab that does. At −110 you need 52.4% just to break even; real edges are small, variance is brutal on small samples, and our own interval math says so. Bet only what you can afford, treat this as research, and if betting stops being recreation, call 1-800-GAMBLER.

What does early access cost?

Early access is free for waitlist members while we finish building. Paid plans come later — the waitlist hears pricing first and locks the founding rate. We’d rather earn the subscription with a season of public grading than charge for promises.

05Early access

On the list before the 2026 season.

Early access opens in waves ahead of the 2026 NFL season. We send one email when your spot is ready — no drip campaigns, no daily “locks.”

One purpose only: telling you when access opens. No spam, no selling your address, unsubscribe any time.