MLB Model Performance Report: July 2026 To Date

Executive Summary

July produced a 74-51 top-five record across 140 published rows, with 12 no-actions and 3 rows still pending or unmatched in the stored data. The Sweat Score 8+ subset finished 74-51, which means the main quality filter remained directionally useful but still needs market-by-market validation before it can be treated as uniform across bet types.

The strongest July conclusions are at the segment level. Markets with larger samples are more informative than one-off perfect records, while small samples are useful only as audit flags for August. No ROI is reported here because the stored daily-card data does not yet contain a complete published-odds field for every pick.

Performance By Market

Market splits show where the published top-five card performed best and where the sample needs more validation.

Performance By Side

Side splits compare over, under, run-line, and other directional calls without mixing them into one black-box record.

Performance By Sweat Score Tier

Score-tier splits audit whether the published score bands separated stronger and weaker rows during July.

Daily Cards Supporting The Totals

ROI, Calibration, And Data Limits

ROI is intentionally not calculated in this report because July daily-card rows do not yet carry a complete, stable published-odds field. Adding ROI before that field is complete would imply precision the source data does not support.

Calibration and Brier-style scoring also remain limited because the stored daily-card article data contains ranks, Sweat Scores, and historical samples, but not a normalized predicted probability for every published pick. The August validation pass should store the probability at publication time so score bands can be audited against realized outcomes.

What Changed And What Comes Next

July included meaningful methodology and presentation changes: daily articles now include richer pick-level analysis, player dashboards expose more initial-HTML history, and results pages show graded records directly to crawlers and users.

For August, the highest-value validation changes are to persist published odds, model probability, model or prompt version, and the exact line source for every top-five row. That will allow ROI, calibration, and version-to-version comparisons to move from qualitative notes into measurable report sections.