The Record

MLB game sims

Our simulation picked the winner in 54.2% of 192 graded games.

Always taking the home team would have won 52.6% of the same games.

Technical score: Brier 0.2495 - lower is better - beats picking by season record (0.2543); does not yet beat always picking the home team (0.2493).

Sample: 192 games, 2026-07-06 to 2026-07-23.

What do these numbers mean?

WNBA v2 game sims

Our simulation picked the winner in 59.2% of 196 graded games.

Always taking the home team would have won 51.5% of the same games.

Technical score: Brier 0.2304 - lower is better - beats always picking the home team (0.2592) and picking by season record (0.2351).

Sample: 196 games, 2026-05-08 to 2026-07-21.

Player prop sims

Over graded props, the player sim scores a Brier score of 0.2012.

First inning (NRFI/YRFI) sim

First-inning model: Brier score 0.2511 vs a flat baseline of 0.2522 over 170 games - narrowly ahead, well short of the 500-game bar.

Today's Tails, self-graded against the closing line

Rolling 30-day window: 95 tails surfaced, 71 graded so far.

60.6% of graded tails beat the closing line, averaging +6.99% closing-line value (CLV).

This is a CLV record, not a win/loss record - beating the closing line is the honest leading indicator that a pick process is sharp; win/loss totals are graded separately per-slip in each member's own log, not aggregated here.

Who grades the slips

33% of graded slips with known provenance were graded automatically by the machine (closing-line grading), not by hand - over 13 graded slips network-wide.

Coming soon

Our own originated lines are being paper-tracked against the close - this section unlocks at n>=100 graded disagreements. No numbers here yet, on purpose.

Winners picked

Out of every graded game our simulation had a live number for, the share where the team it favored actually won.

The 60% test

Calibration: when the model says a team wins 60% of the time, does that group of games actually get won about 60% of the time? A model that says 60% and delivers 60% is honest about its own confidence - one that says 60% and delivers 45% is not.

CLV (closing-line value)

Whether a bet's price was better than the market's final price right before the game started. Beating the close over time is the honest sign of a sharp process, independent of whether any single bet actually won.

The technical score (Brier)

The statisticians' version of the same "winners picked" and "60% test" numbers above, in one number - lower is better. It is published here for anyone who wants to audit the math directly; everything above it says the same thing in plain words.