Alex Bregman (CHC) — actual 44.0 vs proj 9.4 (+34.6)
hitter · slot 2, park×1.05, wx×0.8841
Jonny DeLuca (TB) — actual 28.0 vs proj 3.1 (+24.9)
hitter · slot 6, park×1.04, wx×0.9024
Vinnie Pasquantino (KC) — actual 28.0 vs proj 6.5 (+21.5)
hitter · slot 5, park×1.0, wx×1.004
Daniel Schneemann (CLE) — actual 22.0 vs proj 4.8 (+17.2)
hitter · slot 8, park×1.0, wx×1.004
Christian Walker (HOU) — actual 25.0 vs proj 7.8 (+17.2)
hitter · slot 6, park×0.98, wx×1.0475
Biggest misses (under projection)
Jack Dreyer (LAD) — actual 5.3 vs proj 22.1 (-16.8)
pitcher · outs 18
River Ryan (DET) — actual 2.0 vs proj 17.9 (-15.9)
pitcher · outs 18
Chad Patrick (MIL) — actual -9.7 vs proj 6.1 (-15.8)
pitcher · outs 3
Nick Martinez (TB) — actual -4.8 vs proj 10.9 (-15.8)
pitcher · outs 17.73
JR Ritchie (ATL) — actual 3.4 vs proj 19.0 (-15.6)
pitcher · outs 20.3
Rolling 30-day trend
Single-slate numbers are noisy — this is the honest read. Hitters and pitchers are graded separately.
4.968
hitter MAE (pooled) · r 0.2808 · n 10179
4.547
pitcher MAE (pooled) · r 0.5404 · n 4373
30
graded days
MAE by day (lower better; y-max 6.02)
Correlation by day (higher better; 0–1)
hitters · pitchers
Date
Model
Hitter MAE
Hitter r
Hitter n
Pitcher MAE
Pitcher r
Pitcher n
2026-09-27
v1.1
4.749
0.2699
275
4.651
0.4808
118
2026-09-26
v1.1
5.217
0.3643
268
4.763
0.3666
129
2026-09-25
v1.1
5.174
0.2164
407
5.028
0.6318
168
2026-09-24
v1.1
5.47
0.2951
259
4.439
0.5504
111
2026-09-23
v1.1
4.649
0.2276
494
4.241
0.4452
238
2026-09-22
v1.1
4.922
0.2106
420
4.704
0.5023
189
2026-09-21
v1.1
5.299
0.2598
62
4.84
0.322
22
2026-09-20
v1.1
4.714
0.2645
286
5.035
0.538
116
2026-09-19
v1.1
5.386
0.1792
337
4.415
0.4519
158
2026-09-18
v1.1
4.783
0.3731
473
4.751
0.4456
199
2026-09-17
v1.1
4.836
0.263
162
4.939
0.3688
75
2026-09-16
v1.1
5.022
0.2917
407
4.426
0.4846
176
2026-09-15
v1.1
4.939
0.2978
427
4.212
0.5899
175
2026-09-14
v1.1
4.906
0.3461
238
4.351
0.6437
102
2026-09-13
v1.1
5.099
0.333
271
4.59
0.7098
108
2026-09-12
v1.1
5.061
0.3785
450
4.884
0.5255
185
2026-09-11
v1.1
4.701
0.3738
401
4.652
0.5104
173
2026-09-10
v1.1
4.461
0.235
105
5.148
0.7829
37
2026-09-09
v1.1
5.255
0.2434
281
4.902
0.3839
129
2026-09-08
v1.1
5.071
0.1909
403
3.744
0.5824
172
2026-09-07
v1.1
4.862
0.2972
224
4.722
0.6848
90
2026-09-06
v1.1
4.307
0.4022
341
4.604
0.599
137
2026-09-05
v1.1
4.812
0.2297
343
4.26
0.6755
145
2026-09-04
v1.1
5.2
0.223
504
4.255
0.6089
208
2026-09-03
v1.1
4.318
0.2996
228
4.078
0.6067
97
2026-09-02
v1.1
5.387
0.2013
427
4.453
0.5259
196
2026-09-01
v1.1
5.316
0.2342
439
4.54
0.5201
176
2026-08-31
v1.1
5.267
0.1852
324
4.495
0.6
132
2026-08-30
v1.1
5.029
0.345
280
4.898
0.5892
110
2026-08-29
v1.1
4.708
0.4042
643
4.617
0.4767
302
Live Model
Pregame win-probability and total accuracy from graded snapshots. Confirmed snapshots (lineups + starters posted) are the headline; provisional ones are segregated. Metrics are broken out by model version so pre-fix and post-fix eras never blend. Brier lower is better; skill (vs the 0.5 coin-flip baseline) higher is better; bias is mean signed total error (+ = over-projection).
Model live-2
0.2484
Brier (confirmed) · n 174
0.0064
win-prob skill vs .5
4.082
total MAE
1.145
total bias (signed)
Provisional (n 2): Brier 0.3038 · skill -0.2153 · total MAE 10.236 · bias 10.236
Confirmed — Model vs Market (paired sample, n 160)
Metric
Model
Closing line
Win-prob Brier (n 160)
0.252
0.2482
Total MAE (n 160)
4.063
3.647
Total bias (signed)
1.294
-0.047
Market-relative skill -0.0153 (>0 = model beats the close) ·
Head-to-head: model win prob closer in 48.8%
(78/160 games)
Line movement (CLV)
Win prob (model↔open disagreement ≥ 3.0 pp; n 110):
close moved toward model in 34.5% ·
mean |open−close| 0.0115 vs mean |model−close| 0.0851
Total (model↔open disagreement ≥ 0.5 runs; n 119):
close moved toward model in 12.6% ·
mean |open−close| 0.1261 vs mean |model−close| 1.7399
Excluded 9 cold-start game(s) — a single pregame line (open = close), no movement to measure.
Provisional — Model vs Market (paired sample, n 2)
Metric
Model
Closing line
Win-prob Brier (n 2)
0.3038
0.2968
Total MAE (n 2)
10.236
9.25
Total bias (signed)
10.236
9.25
Market-relative skill -0.0236 (>0 = model beats the close) ·
Head-to-head: model win prob closer in 50.0%
(1/2 games)
Line movement (CLV)
Win prob (model↔open disagreement ≥ 3.0 pp; n 1):
close moved toward model in 100.0% ·
mean |open−close| 0.0134 vs mean |model−close| 0.0489
Total (model↔open disagreement ≥ 0.5 runs; n 1):
close moved toward model in 0.0% ·
mean |open−close| 0.0 vs mean |model−close| 1.1574
Excluded 1 cold-start game(s) — a single pregame line (open = close), no movement to measure.
Model live-3
0.2448
Brier (confirmed) · n 926
0.0207
win-prob skill vs .5
3.487
total MAE
-0.222
total bias (signed)
Provisional (n 1): Brier 0.1639 · skill 0.3445 · total MAE 8.548 · bias 8.548
Confirmed — Model vs Market (paired sample, n 862)
Metric
Model
Closing line
Win-prob Brier (n 862)
0.2456
0.2398
Total MAE (n 862)
3.482
3.374
Total bias (signed)
-0.217
-0.526
Market-relative skill -0.0242 (>0 = model beats the close) ·
Head-to-head: model win prob closer in 46.1%
(397/862 games)
Line movement (CLV)
Win prob (model↔open disagreement ≥ 3.0 pp; n 548):
close moved toward model in 41.1% ·
mean |open−close| 0.0099 vs mean |model−close| 0.0831
Total (model↔open disagreement ≥ 0.5 runs; n 471):
close moved toward model in 11.9% ·
mean |open−close| 0.1221 vs mean |model−close| 1.1393
Excluded 51 cold-start game(s) — a single pregame line (open = close), no movement to measure.
Provisional — Model vs Market (paired sample, n 1)
Metric
Model
Closing line
Win-prob Brier (n 1)
0.1639
0.2722
Total MAE (n 1)
8.548
7.5
Total bias (signed)
8.548
7.5
Market-relative skill 0.3979 (>0 = model beats the close) ·
Head-to-head: model win prob closer in 100.0%
(1/1 games)
Line movement (CLV)
Win prob (model↔open disagreement ≥ 3.0 pp; n 1):
close moved toward model in 100.0% ·
mean |open−close| 0.0053 vs mean |model−close| 0.1169
Total (model↔open disagreement ≥ 0.5 runs; n 1):
close moved toward model in 0.0% ·
mean |open−close| 0.0 vs mean |model−close| 1.0478
Market comparisons use vig-removed lines (two-sided moneylines normalized to sum to 1), on the paired sample only — snapshots with no captured odds are excluded, never zero-filled.