by bpleone

About EdgeStat

EdgeStat is the public face of a private quantitative sports-betting research desk run by Brandon Leone. Built in Southern California. Powered by Python, Postgres, and a lot of cold coffee.

Our Mission

Most sports bettors lose, and the industry profits from that fact. We think this is largely an information problem. Books have invested in modeling for a decade; bettors haven't. EdgeStat is built to bridge that gap — to give recreational bettors the same quantitative tools the books use, so they can compete on a level field.

We're a research operation first, a publication second. Our outputs — the daily slate, Play of the Day, sharp flow alerts, the Academy — are how we share what we find. The numbers behind every recommendation are the same numbers we use ourselves.

Founder

Brandon Leone is a former financial analyst (IB / equity research) based in Southern California. He spent half a decade modeling equities and cash flows on Wall Street's clock; now he models baseball games and player props on his own.

If the underlying math of "what's this company worth?" feels eerily similar to "what's this team's true win probability?" — that's because it is. Both are about pricing uncertainty correctly. The sports market is younger, noisier, and less efficient — which is exactly why there's still alpha here.

Reach out: brandonpleone@gmail.com

Our Process

Every morning at 6:00 AM ET the data pipeline runs autonomously on GitHub Actions. It (1) settles yesterday's projections against final box scores from MLB Stats API, (2) recomputes per-market calibration metrics on the trailing 30 days, (3) pulls today's schedule, lineups, weather (Open-Meteo), home-plate umpire, and DraftKings odds (The Odds API), (4) layers Baseball Savant Statcast metrics (xK%, xBA, xSLG, barrel%), (5) writes today's fair prices for moneyline, total, and the four player-prop markets, plus a side-by-side PrizePicks comparison.

Two more refreshes at 1 PM ET and 6 PM ET catch line moves and confirmed lineups. The parlay engine combines highest-edge legs into 2- and 3-leg suggestions with same-game correlation haircut. The model bias-corrects per market once 30+ outcomes have flowed -- so day 30 trusts the data; day 1 trusts the priors with caps.

After every game settles, results append to track-record.json (in git history -- no revisions, no cherry-picks). Calibration runs nightly. Brier, log loss, ECE, and ROI roll forward continuously. The model only gets sharper.

What we are not

Roadmap

Last updated: May 2026.