How the 2026-09-21 predictions scored against actual results
Cumulative Δlog-loss vs always-pick-home: -0.0243 ± 0.0068 per game over 384 games (paired; negative = beating the baseline - currently ahead).
| Yesterday | Correct | Log-loss |
|---|---|---|
| model | 3/3 | 0.606 |
| always-pick-home | 3/3 | 0.627 |
Daily detail: 100% hit rate · Brier 0.207 · avg margin error 3.4 · avg total-runs error 2.0. Running: 230/384 (59.9%) · cumulative log-loss 0.666 · cumulative Brier 0.237
| Game | Pick | Predicted | Actual | |
|---|---|---|---|---|
| ✅ | Blue Jays @ Orioles | Orioles (51%) | 4.5–4.4 | 4–3 (home first) |
| ✅ | Nationals @ Tigers | Tigers (58%) | 4.8–4.2 | 9–2 (home first) |
| ✅ | Twins @ Giants | Giants (54%) | 4.7–4.5 | 5–2 (home first) |
The headline is the paired per-game log-loss difference against a fixed always-pick-home forecast (p=0.534) on the same games, ± one standard error; daily hit rate is noisy and demoted on purpose. Reference points: 0.693 is a coin flip; the model's walk-forward backtest runs 0.680 (research/SP-BACKTEST.md).
Betting-blind Elo model (K=3, +24 home, MOV-weighted, 60% season carryover) · Can Tre Beat Vegas · whosyurgoat.app/vegas/mlb
This email online: 2026-09-21 grade · Previous: 2026-09-20 · All MLB emails