LIVE DAILY BOARD · JULY 20, 2026

Two picks today.
Marte + Perdomo.

DOUBLE DOWN — Ketel Marte (0.733) + Geraldo Perdomo (0.691).

Generated July 20, 2026 at 8:16 PM CT · 270 of 270 batters from announced lineups · picks lock at first pitch · read the case study

Active streak

2

consecutive winning picks · since Jul 19

Pick record

12–4

every pick played · 8 days

Top pick win probability

73.3%

Ketel Marte · AZ

2025 validation

74.7%

daily top pick · 182 days, walk-forward

THE BOARD

Every pick, ranked.

Recalibrated P(≥1 hit) for every projected starter today. The play policy takes the top two when both clear 66%; otherwise one pick, or none at all. Picks lock at first pitch; statuses and box lines update live.

PlayerGame Win probability
1James Wood#1 WSH @ COL · vs Kyle Freeland announcedLIVE73.9%
2CJ Abrams#4 WSH @ COL · vs Kyle Freeland announcedLIVE73.3%
3Curtis Mead#2 WSH @ COL · vs Kyle Freeland announcedLIVE73.3%
4Ketel MartePICK#1 AZ vs ATH · vs Jeffrey Springs announced8:40 PM CT73.3%
5Vladimir Guerrero Jr.#3 TOR vs TB · vs Nick Martinez announcedLIVE73.3%
6Luis Arraez#1 SF @ KC · vs Michael Wacha announcedLIVE72.7%
7Jake McCarthy#1 COL vs WSH · vs Andrew Alvarez announcedLIVE71.8%
8Bobby Witt Jr.#2 KC vs SF · vs Trevor McDonald announcedLIVE69.9%
9Dylan Crews#5 WSH @ COL · vs Kyle Freeland announcedLIVE69.1%
10Ceddanne Rafaela#2 BOS vs BAL · vs Shane Baz announcedLIVE69.1%
11Ernie Clement#1 TOR vs TB · vs Nick Martinez announcedLIVE69.1%
12Jeremy Peña#1 HOU vs MIA · vs Janson Junk announcedLIVE69.1%

TRACK RECORD

Watch it calibrate.

The model's numbers are only worth what they predict. Every probability below is out-of-time: fit on the past, scored on what came next. Live results join the chart as they settle.

Predicted vs. actual

2025 walk-forward season, 41,117 picks in 12 buckets

40%50%60%70%80%40%50%60%70%80%perfect calibration
View as table
PredictedActualPicks
24.0%23.8%3,463
43.4%43.2%3,468
52.3%52.8%3,383
56.2%55.6%3,463
59.2%59.8%3,419
60.3%59.6%3,363
62.7%63.5%3,478
64.4%64.5%3,652
65.7%65.9%3,349
66.2%65.4%3,243
67.9%67.5%3,437
70.9%70.1%3,399

Daily top pick, 30-day rolling

Hit rate of the #1 board pick

2025 validationLive
60%70%80%90%Apr ’25Jun ’25Sep ’25live
View as table
Date30-day hit rate
2025-04-0520.0%
2025-04-1958.3%
2025-05-0373.3%
2025-05-1776.7%
2025-05-3176.7%
2025-06-1473.3%
2025-06-2880.0%
2025-07-1276.7%
2025-07-3066.7%
2025-08-1366.7%
2025-08-2786.7%
2025-09-1090.0%
2025-09-2480.0%

THIS SEASON

The streak, day by day.

Every settled pick, in order. The line climbs one step per winning pick and drops to zero the moment one goes hitless: the same 12-for-16 record and 2-streak on the tiles above, drawn out over 9 settled days. The line is riding a 2-pick streak, live since Jul 19.

streak lengthstreak broken
012345Jul 6Jul 8Jul 9Jul 16Jul 17Jul 192
View as table
DatePicksStreak
Jul 6✓ Shohei Ohtani 72%; ✓ Luis Arraez 70%2
Jul 7✓ Shohei Ohtani 72%; ✓ Fernando Tatis Jr. 71%4
Jul 8— Byron Buxton 72%; ✓ Otto Lopez 72%5
Jul 9✗ Gunnar Henderson 72%; ✗ Trea Turner 72%0
Jul 10✓ Luis Arraez 73%; ✓ Bobby Witt Jr. 72%2
Jul 16✗ Bo Bichette 69%0
Jul 17✓ Jake McCarthy 73%; ✓ Fernando Tatis Jr. 73%2
Jul 18✓ Elly De La Cruz 74%; ✗ Sal Stewart 73%0
Jul 19✓ Elly De La Cruz 73%; ✓ Ketel Marte 73%2

PICK LOG

Every day, on the record.

Picks are logged before first pitch and never edited. Outcomes settle the next morning when official data posts.

Mon Jul 20···James Wood 73.9% — pending···CJ Abrams 73.3% — pending
Sun Jul 19Elly De La Cruz 73.3% — hitKetel Marte 73.2% — hit
Sat Jul 18Elly De La Cruz 73.9% — hitSal Stewart 73.3% — no hit
Fri Jul 17Jake McCarthy 73.2% — hitFernando Tatis Jr. 73.2% — hit
Thu Jul 16Bo Bichette 68.8% — no hitTrea Turner 66.0% — hit
Fri Jul 10Luis Arraez 72.6% — hitBobby Witt Jr. 72.2% — hit
Thu Jul 9Gunnar Henderson 72.2% — no hitTrea Turner 72.2% — no hit
Wed Jul 8Byron Buxton 72.4% — did not playOtto Lopez 72.4% — hit
Tue Jul 7Shohei Ohtani 72.0% — hitFernando Tatis Jr. 71.4% — hit
Mon Jul 6Shohei Ohtani 72.1% — hitLuis Arraez 70.0% — hit

METHOD

Built to be doubted.

Walk-forward, no leakage

Every feature is computed strictly from games before the prediction date. The model retrains every morning and never sees the day it predicts.

Ideas earn their way in

Model changes must beat the incumbent on three held-out seasons. Eight upgrades failed that gate and were rejected. The receipts are in the case study.

Externally checked

On matched player-days, the model's probabilities price hits at parity with professional sportsbook lines, built only from free public data.

Honest about the odds

A great top pick wins about 4 days in 5. Misses are part of the math. The streak comes from compounding good probabilities, not certainty.

Read the full case study