How It Works

Why this page exists

Most prediction services start their track record on their best week and quietly drop the picks that would ruin the average. A win rate with no record behind it is just a claim.

Sharp Retriever runs on a different standard: every play is logged before the game, graded on the final score, and the record can't be edited. That's how a sharp grades himself. It's how we grade ourselves.

The model reads the game.

Every MLB matchup gets scored by our model before first pitch. It's an XGBoost classifier trained on years of Retrosheet game data, reading the stuff that actually decides baseball games — starting pitcher matchup (xwOBA, swinging-strike rate, contact quality), bullpen state and rest, team batting splits by handedness, park factors, weather, umpire. What it does NOT read: the betting market. We don't strip vig, we don't compare sportsbook prices, we don't chase line movement. The market is not an input. The read comes from baseball, or it doesn't come at all.

The pick publishes before first pitch.

When the model has a read, it publishes — team, market, and a tier: Big Dog, Bird Dog, or Guard Dog. Tiers tell you how the model sees the shot, not how sure we are you'll cash. Every read ships with the reasoning behind it, in plain language. It's locked before the game starts and it never moves. No edits, no takedowns, no quiet walk-backs.

The record keeps itself.

Every pick gets graded on the final score and written to a record that can't be edited — not by us, not by anyone. Wins, losses, pushes, all of it, out in the open. We don't hide a bad week and we couldn't if we wanted to; the ledger is append-only and enforced at the database level. Anyone can see how every tier and every market has actually done. That's the whole pitch: we're not selling you a lock, we're showing you the truth and daring you to catch us bending it.

Every read hits your dashboard with the full reasoning behind it, the moment it publishes. Nothing held back.