Model History
Every graded NYX prediction — the record, the trend, and how honestly the model states its confidence.
SEASON RECORD
352–306
658 graded picks
SEASON ACCURACY
53.5%
+3.1 pts vs first 14 days
LAST 7 DAYS
44–36
55.0% hit rate
MODEL CONFIDENCE
55.4%
avg stated win% · max 93%
ACCURACY TREND14-day rolling · Jun 26 – Aug 10
Rolling accuracyCoin flip (50%)
Best stretch Jul 14–20 · 66.0%Worst stretch Jul 10–16 · 46.2%Each point is a trailing 14-day window of graded picks.
MONTHGAMESMODEL RECORDACCURACYBRIER
June (from 13th)18394–8951.4%0.247
July342182–16053.2%0.251
August (through 10th)13376–5757.1%0.244
Season658352–30653.5%0.248
Brier score — squared error of the stated win% · lower is better · 0.250 matches a coin flip.
Record by Model Confidence Higher-confidence calls should hit more often — and they do.
Strong 70%+ stated win%5–3 · 62.5%
Moderate 60–70%75–53 · 58.6%
Slight 50–60%272–250 · 52.1%
Graded picks only · ties and postponed games excluded.
When NYX says a team wins X% of the time — how often it actually did.
50–55%49.8%n=402
55–60%60.0%n=120
60–70%58.6%n=128
70%+62.5%n=8
Stated win rates run hot by ~4.8 pts — actual wins land below the stated number.
Newest settled games first.
DATEMATCHUPNYX LEANCONFRESULT
Aug 10Rockies @ D-backsD-backs56%HIT
Aug 10Rays @ AthleticsAthletics52%MISS
Aug 10Mets @ BravesBraves52%MISS
Aug 10Rangers @ AngelsAngels52%MISS
Aug 10Royals @ DodgersDodgers60%HIT
Aug 10Orioles @ TwinsTwins52%HIT
Aug 10Brewers @ PadresPadres52%HIT
Engine v2.2 · calibration refit daily · Graded picks settle the morning after each final · NYX reports win probabilities — a model read, nothing more.