Sharp Retriever

About

A prediction model built to be honest.

What we are

Sharp Retriever is a subscription-based MLB prediction product. We build machine-learning models that predict game outcomes — moneyline, totals, run lines — from baseball signal: starting-pitcher quality, bullpen fatigue, park factors, umpire tendencies, weather, and lineup matchups. We package those predictions as daily picks and publish them to subscribers before first pitch.

What the product is not

We do not sell price-vs-price arbitrage or "value" derived from comparing one sportsbook to another. The model's picks come from baseball signal — the market line is not an input feature. We believe that approach is more durable: the signal either knows something the market doesn't, or it doesn't, and the record will prove which.

How it works

Each morning, the pipeline assembles game-day features — starter xwOBA, CSW%, swinging-strike rate, bullpen workload over the last 1/2/3 days, umpire run-scoring tendency, park run factor, and weather environment — and runs them through an XGBoost model trained on 2015–2024 Retrosheet results. The model outputs a probability for each side. Model picks are published to the slate. Every pick is locked before first pitch; we do not adjust published picks after the game starts.

How we track the record

Every pick is logged with the market odds at publish time. After the game settles, the result — win, loss, or push — is recorded against the locked pick. Nothing is held back, cherry-picked, or backdated. The forward record begins from day one of publication.

Who runs this

Sharp Retriever is operated by a single person. The model is built, maintained, and published without a team. That means slow iteration by design — we add signal only when it's verified, and publish results only when the sample is meaningful. No hype, no fake consensus, no testimonials. The record stands on its own.

Need help? Contact us — we respond within 24 hours.