CourtVision / forecast calibration

How forecasts
compare with outcomes.

These MLB/soccer numbers are revision 2, computed on the segment-clean corpus (2026-09-16). Revision 1 values are withdrawn and kept in the regeneration receipt. Read the finding.

For MLB, model forecasts of 50% to 60%, the mean forecast was 54.33% and the observed frequency was 55.20%, from 5,636 observations across 114 games. Snapshot date: 2026-09-16.

Forecast mean54.33%

Observed frequency55.20%

Observed-frequency interval: 44.03% to 66.70% (95.00% game-cluster bootstrap interval; 2.5% to 97.5% percentiles; 1,000 resamples).

The published gap is 0.87 percentage points; its interval is -10.26 to 12.17 percentage points, includes zero, and does not establish agreement.

Inspect this exact calibration bin

Published calibration snapshot.

Published snapshot 2026-09-16

Probability quality

Model vs. market

Source

CourtVision modelMarket baseline

Baseball Game winner27,351 rows

Model

0.1668

Market

0.1510

MLB in-game moneyline (game winner). Shared axis 0-0.50.

Soccer Game winner4,265 rows

Model

0.3325

Market

0.1952

soccer in-game match result (2-way, as graded). Shared axis 0-0.50.

Baseball Totals2,113 rows

Not scoredno resolved outcome label on rescued live quote-tracks (settlement join unavailable)

Brier measures squared probability error, including calibration and resolution. Rows can include multiple observations of a game; they are not independent games. These comparisons describe published probability error only.

Inspect every reliability bin

Monthly snapshots

How quality changes over time

Source

Same metric, separate sport cohorts. Only the 2 published months are shown.

Baseball

Mlb monthly brier measurements
Month	Model	Market	Rows
2026-06	0.1968	0.1873	5,030
2026-07	0.1600	0.1428	22,321

Soccer

Soccer monthly brier measurements
Month	Model	Market	Rows
2026-06	0.3273	0.1955	1,213
2026-07	0.3346	0.1950	3,052

2026-06 to 2026-07. Changes in cohort composition can change these scores. Connecting observations is a visual guide, not an interpolated estimate.

Research records

Published research outcomes

Source

Confirmed groupNull / otherNot testable

Basketball85 records

45337

Baseball86 records

32468

Soccer64 records

29278

Tennis52 records

24208

Grouped as recorded in the ledger. Local confirmation is not proof of causality. Open records to distinguish provisional and null findings.

Derived analyses

Recently added analyses

Calibration by game checkpoint

All sports / 16 rows

Calibration by published probability band

All sports / 10 rows

Brier skill score by sport and game phase

All sports / 11 rows

Signed calibration direction

All sports / 4 rows

Cluster-bootstrap uncertainty width

All sports / 4 rows

Calibration support concentration

All sports / 4 rows

Analysis pages

Related analysis pages

Experimental measurements

9 experimental measurements for lineups, fatigue, forecasts, and dependence.

Published findings

Measured results with their limitations and supporting evidence.

Player comparisons

Statistical profiles and available percentile comparisons.

Research checks

Testing steps and their recorded outcomes.

Baseball / 2025 Statcast

Inside the pitch corpus

Source

693,037pitches in the published pull

Four-seam fastball

220,235 pitches in mix

P1091.3 mph

Median94.5 mph

P9097.7 mph

Velocity observations: 220,233. Counts differ when velocity is missing.

Nine most common pitch types shown. Descriptive pitch mix and velocity; not a predictive result.

Basketball / coverage audit

How complete are player dossiers?

Source

18 of 28median categories filled: 18 of 28 (64.3%)

46.4%published completeness score: 46.4%

1,249dossiers audited

The artifact does not define how the published completeness score is derived.

Categories filled per dossier

0 categories1 dossiers (0.1%)

1 categories77 dossiers (6.2%)

2 categories70 dossiers (5.6%)

3 categories173 dossiers (13.9%)

4 categories100 dossiers (8.0%)

5 categories8 dossiers (0.6%)

6 categories9 dossiers (0.7%)

7 categories16 dossiers (1.3%)

8 categories22 dossiers (1.8%)

9 categories15 dossiers (1.2%)

10 categories22 dossiers (1.8%)

11 categories19 dossiers (1.5%)

12 categories8 dossiers (0.6%)

13 categories11 dossiers (0.9%)

14 categories12 dossiers (1.0%)

15 categories5 dossiers (0.4%)

16 categories10 dossiers (0.8%)

17 categories20 dossiers (1.6%)

18 categories82 dossiers (6.6%)

19 categories72 dossiers (5.8%)

20 categories22 dossiers (1.8%)

21 categories71 dossiers (5.7%)

22 categories61 dossiers (4.9%)

23 categories179 dossiers (14.3%)

24 categories142 dossiers (11.4%)

25 categories22 dossiers (1.8%)

Category fill rates

Rebounding Profile95.4%

Foul Drawing86.6%

Spacing Gravity78.4%

Durability Load61.5%

Catch Shoot Vs Pullup57.8%

Playmaking Network57.8%

Rest B2b Splits57.8%

Shot Clock Scoring57.8%

Usage Role57.8%

Shot Profile57.5%

Form Streak Dynamics56.5%

Vs Scheme Splits56.1%

Situational Splits55.2%

Transition Scoring53.7%

Monthly Form51.2%

Defensive Profile50.5%

Post Up Profile46.3%

Pick And Roll Profile46.1%

Ft Profile45.7%

Scoring Creation44.8%

Score Margin Splits43.2%

Quarter Shape Fatigue39.8%

Isolation Profile39.4%

Foul Tendency37.2%

Clutch Scoring35.1%

Matchup Splits8.5%

Turnover Profile1.4%

Pace Fit1.0%

Fill rate is the fraction of dossiers containing a category. Presence does not establish accuracy.

Derived analyses69Interactive analyses from published snapshots

Snapshot checks96/96Recorded in the published manifest

This workspace reads published snapshots. Dates and populations vary by source. Snapshot checks describe recorded results, not current system health.

Methodology