Your paper bets ticked on the day pages, kept in this browser, scored when games go final
Model strategies at fair odds 47 graded MLB days · flat 1u · fair = the odds the model's own % implies, so a real book pays less than this
Strategy
Bets
Record
Hit %
Model said
Units
ROI
Calibration does 25% mean 25%?
Model %
n
Predicted
Actual
Diff
Raw said
The rigged test, Kalshi money hitters with the most public dollars on them: do they underperform?
Kalshi volume
n
Model said
Crowd said
Actual
Actual ÷ crowd
The rigged test, heat do the crowd's names underperform their own numbers?
Heat
n
Predicted
Actual
Actual ÷ predicted
Picks are locked before first pitch and graded from box scores afterward; results never change a lock. Days before 2026-09-03 were reconstructed from posted lineups with season stats as of the build, which leaks a little. Model v3 (from 2026-09-19): home-run percentages are recalibrated because v2 was too spread out (hitters shown at 30% homered 22%, hitters shown at 4% homered 7%); tested on held-out days, and the table above scores the two eras separately, with Diff in green when predicted and actual are within 2 points. Model v2 (from 2026-09-12): base rate regressed less toward league, weak hitters dampened, level scaled 0.92 - fitted on those same days, so judge it on days after that.