Receipts · Public Record

The Ledger

Anyone can call a game after it happens. We put every number on the record before — snapshotted the moment it's published, graded against the result, and never backdated. This page is the running scorecard. Record began Jul 11.

Biggest moves

Movement shows up once a metric has two checkpoints on the record. The next MLB refresh appends tonight — the first movers land then.

Forecasts on record

2,215 snapshots

MLB

1,050

30 teams · 1 checkpoint

Last logged Jul 15

NBA

450

30 teams · 1 checkpoint

Last logged Jul 11

College Football

272

136 teams · 1 checkpoint

Last logged Jul 11

Premier League

200

20 teams · 1 checkpoint

Last logged Jul 13

NFL

128

32 teams · 1 checkpoint

Last logged Jul 11

WNBA

60

15 teams · 1 checkpoint

Last logged Jul 15

F1

44

22 teams · 1 checkpoint

Last logged Jul 15

F1 Constructors

11

11 teams · 1 checkpoint

Last logged Jul 15

MLB game predictions

the board →

Before each game's first pitch, the model locks in a win probability. Once the game is final, we score it — no edits, no hindsight. Here's how the calls have held up.

New here? What these numbers mean
Right side —
how often the team we favored actually won. Just win/loss on our picks.
Brier score —
our grade for the probabilities themselves, not just which side we picked. It rewards being right and being honest about how sure we were: a confident call that hits scores well, a confident call that misses gets punished hard. Saying 50% on everything (a coin flip) scores 0.25 and a flawless forecaster scores 0 — so the lower the number, the sharper our odds have been.
Calibration —
whether our confidence is honest: when we say a team has a 70% chance, do teams like that really win about 70% of the time?

Games called

29

locked before first pitch

Now final

28

played & scored

Right side

61%

our pick won 17 of 28

Brier score

0.258

forecast accuracy · 0.25 = coin flip

Calibration

predicted vs actual

When the model gives a team, say, a 60% chance — do teams like that actually win about 60% of the time? Each row is how often our picks in that range really won.

40–50%
30%
50–60%
60%
60–70%
33%

actual predictedWell-calibrated → the marker sits at the edge of the bar.

Recently graded

Locked predictions

home win prob · graded when final

The bolded team is the model's pick. Each line is locked before first pitch and graded automatically once the game is final.

Methodology: MLB home-win probabilities come from a log5 matchup on each club's regressed Pythagorean rating plus a fixed home-field edge — the same engine that runs the season simulation. Predictions are written before first pitch and cannot be edited. Brier score is the mean squared error against the 0/1 outcome; 0.250 is a coin flip, lower is better. Nothing on this page is backdated.