About Sherlock Picks
A sports picks platform built around transparent results, readable analysis, and community support.
Our Mission
Sherlock Picks is built to make sports modeling easier to understand. The goal is not to promise wins. The goal is to show probability, price, and context clearly enough that members can make their own decisions.
Every model is checked with walk-forward backtesting so each fold is evaluated on games the model had not seen during training. The probability calibrator that ships with each model is then fit on the stacked out-of-fold predictions from every walk-forward fold — not on an in-sample slice of the training data — so posted probabilities reflect genuinely out-of-distribution behavior. We also track calibration, expected value, and historical behavior so performance can be judged with context instead of hype.
How It Works
Collect the data
Historical results, upcoming schedules, injuries, market odds, and league-specific stats are gathered and cleaned.
Test the models
Models are backtested with rolling walk-forward folds, then calibrated on the out-of-fold predictions collected across those folds, and checked against quality metrics.
Publish with context
Predictions, reports, and supporting pages are synced into the site with feature access based on Trial, Supporter, and Lifer roles.
Current Coverage
The live system currently supports 8 sports in the core pipeline (NBA, NFL, MLB, NHL, UFC, WNBA, College Football, Men's College Basketball), plus NBA player props and an NBA first-basket module. Soccer coverage is paused pending a historical data gap resolution.
What powers the site
How the project keeps moving
Sherlock Picks is funded by supporters. Tips help cover hosting, data operations, and new feature work. Current priorities include better calibration, stronger prop coverage, and clearer in-app performance evidence.