MLB Picks Today,
with the losses still attached.

Every candidate market on the slate gets a model probability first: player-level inputs for props, team and matchup context for game lines. That probability is then compared against a no-vig consensus price — the market with the sportsbook margin stripped out — so an edge is measured against what the number should be, not against a single book's markup. Only the gap that survives both checks is eligible to become a pick, and the full decision set (BET, WAIT, PASS) is frozen into an immutable lock archive before first pitch.

Qualified picks are graded by the size and quality of the edge, not by confidence theater. STRONG is the higher band; LEAN still cleared every gate with less room to spare. Rows that fail — stale quotes, thin markets, missing lineup confirmation, an edge too small to survive the vig — are held at WAIT or declined at PASS, and those decisions are archived too. The count on this page is the number that survived, which is why some slates publish a board with zero picks. Grades are never upgraded after the fact: a LEAN that wins was still a LEAN at lock, and the archive keeps the original band so the tier record can be audited on its own.

After settlement, every row flows into the delayed public ledger at /data with its model probability, market reference, grade, and outcome — losses included. Season aggregates live at /mlb/accuracy, and the reasoning behind the gates is written up at /methodology. The record is delayed, but it is never curated.

Baseball changes after lock: a hitter is scratched, a starter is bumped, a game is postponed. Posterior does not delete those rows — they settle as voids with the reason attached, whether the lineup was never confirmed or the player was pulled late. Voided picks are excluded from win rate and ROI but kept in the ledger and counted in the board's assessed total, so the day's full decision set stays reconstructable: what was bet, what was held, what was declined, and what baseball cancelled.