Sports Modeling Hub

Let's beat Vegas together

  • No code
  • Custom ML models
  • Backtested on real seasons
Start building, free

Not a sportsbook. No money is placed on SMH.

Skip to the finished example
Example model, illustrative numbersNov 17, 2026
Lead backRushing yardsvs. example defense, Week 12
68Model score
  • Recent form onHis last few games
  • Matchup onHow well the defense stops the run
  • Game script onTeams that lead run late
  • Weather onRain and wind mean more runs
Predicted rushing yardsAbout 79 yards
Beats the plain average by 7Baseline 61
Trained on 136 gamesGradient-boosted model
Example model, illustrative numbersNov 17, 2026

How this score was made

How much each signal mattered
  • Recent form38%
  • Matchup27%
  • Game script20%
  • Weather15%
Method
Gradient-boosted trees. Random forest and ridge were tried too.
Tested on
136 past games it never saw while training
Baseline
His average over the last 5 games
Calibration
Its ranges held the real result about as often as they claimed

Past results do not predict future ones.

Build a model in four steps

Pick what you want to predict.

Start with one stat. This example predicts rushing yards: how far a running back carries the ball in a game.

Add the signals you believe in.

Recent form, the matchup, how the game is likely to go, the weather. Switch each one on and watch the score move.

Train it on real plays.

SMH reads past seasons of public play-by-play data and fits the model for you. No code, no spreadsheets.

Check it against a plain average.

The baseline is his recent average. This example scores 68 against the baseline's 61, so its signals add something. If yours don't, you'll know.

That's a model.

Flip the card to see what it learned and how it was tested. Then build your own.

Start building, free

Build a model without code.

Choose a stat, switch on the signals you trust and pick how the model learns. SMH handles the data.

  • Rushing, receiving and passing yards, receptions and more
  • Signals in plain words, each with a one-line explanation
  • Gradient-boosted, random forest or ridge models
Example model, illustrative numbers

What to predict

Signals

Recent formMatchupGame scriptSnap share

How it learns

Learns from combinations of signals. A good default.

See how it would have done.

The backtester replays past weeks your model never saw and checks each guess against the recent average. Some models miss. SMH shows you which.

  • Week by week, model against the plain average
  • Misses are marked, not hidden
  • Calibration shows whether its ranges hold up
Example model, illustrative numbers. Weeks 1 to 11, 20260102030401234567891011Week 7: missed
Backtest of the example model, week by week.
Your modelRecent averageMissed the average
Average miss, model 13 ydsRecent average 17 yds1 week worse than average

Find where your model disagrees.

Bet Finder puts your model's number next to this week's line for every player, so the biggest gaps are easy to spot.

  • Search any player or stat
  • Sorted by how far your model is from the line
  • Just numbers. Nothing to click and wager
Example data, illustrative numbers. Week 12
  • Starting QB (example)Passing yards
    Your model231 ydsThis week's line248.5
    17.5 below the line
  • Lead back (example)Rushing yards
    Your model84 ydsThis week's line72.5
    11.5 above the line
  • Slot receiver (example)Receiving yards
    Your model61 ydsThis week's line58.5
    2.5 above the line

Showing 3 players

Track picks with your reasons.

Write down what you think and why, on paper or for a pick you made somewhere else. Come back after the game and see who was right.

  • Paper picks or real ones, kept apart
  • Your reasoning saved with every entry
  • Results filled in after the game
Sun, Nov 23

Lead back, rushing yards. I think more than 72.5.

Their run defense has slipped, and we should be ahead late. No rain.

Result: fills in after the game

SMH never takes a bet. The tracker is a notebook, not a betting slip.

Share models. Argue about Sunday.

See how other fans built their models and why. Join a league with friends, ranked by points, never money.

  • A feed of shared models with the reasoning behind them
  • A game thread for each week's lines
  • Copy any shared model as your starting point
R
rushing_rita shared a modelExample post

Rain games, run game. In November I weight weather and game script above recent form. Scores 66 against a baseline of 61.

Week 12 thread: lead back rushing yards

W

weather_wes Wind is up to 20 mph. My model says 81.

L

lineman_leo Their front is healthy again. Mine says 70.

Example league standings

  1. 1rushing_rita412pts
  2. 2You388pts
  3. 3lineman_leo361pts

Tuesday Film Club, example league

Earn points, badges and daily missions.

Points come from building, testing and sharing. They are for bragging rights and league standings, nothing else.

  • Short daily missions to learn the tools
  • Badges for milestones, like beating the baseline
  • Points can't be bought, cashed or spent

Today's missions

388 season points
  • Switch on a new signal (done)+50
  • Track a pick with a reason (done)+20
  • Reply in a game thread (to do)+10
First model
Beat the baseline
Five picks tracked
Shared a model

A notebook, not a betting window.

Not a sportsbook. No money is placed on SMH. If gambling is a problem for you or someone you know, call 1-800-GAMBLER. Read our Responsible play page.

  • We don't accept wagers.No bets are placed on SMH, ever.
  • We don't hold funds.No deposits, balances or payouts.
  • We don't link to sportsbooks.No offers, promos or referrals.
  • Models can be wrong.A score is an estimate from past games, not a promise. Past results do not predict future ones.

Quick answers.

Is this a sportsbook?

No. Not a sportsbook. No money is placed on SMH. It's a modeling and tracking tool: we don't accept wagers, hold funds or link to sportsbooks.

Do I need real money?

No. Track picks on paper, or log ones you made elsewhere. Nothing on SMH asks for money.

Do I need to code?

No. You pick a stat and switch signals on. SMH does the data work using public play-by-play data.

What is the model score?

A 0 to 100 grade for how well your model predicted past games it didn't train on, shown next to a plain recent-average baseline. It's an estimate and can be wrong.

Is it free, and how do I sign in?

It's free to start. You sign in with Google today; more sign-in methods are coming.

Your read on Sunday, with the receipts.

Start building, free

Not a sportsbook. No money is placed on SMH.