MATCHDAY MODEL RECAP
Matchday's model graded 53 MLB picks this stretch, hitting on 30 of them (57%).
On calibration: picks stated at 45-55% hit 71% of the time (7 games); picks stated at 55-65% hit 58% of the time (12 games); picks stated at 65+% hit 53% of the time (34 games). A well-calibrated model's stated confidence should roughly match its actual hit rate in each band.
The strongest signal lately has been season record: when it favored a side, that side won 66% of the time (29 games). Elo rating has been the weakest, at 49% (37 games) — a reminder no single factor is decisive on its own.
Among this week's MLB storylines: the biggest upset was Houston Astros winning 5-9 against Chicago White Sox (the market gave that result only about 0%); the model's best call was Chicago Cubs in Pittsburgh Pirates vs Chicago Cubs at 61% confidence; its biggest miss was picking Boston Red Sox in Oakland Athletics vs Boston Red Sox, but Oakland Athletics won instead.
This is Matchday's own model reporting on itself — not betting advice. Every pick is locked before kickoff and graded automatically once the game finishes; none are rewritten after the fact. See the Q&A page for how the model works.
Matchday — pregame predictions, market context and postgame grading.