The box score is the last place a breakout shows up.
Gavin Williams's contact quality moved before his slash line did. The gap between the two is what we flag — and today it is the widest on the board.
+21 pctile · cohort standing
Player-level readings render live from the calibrated model and change nightly. The published, measured figures on this page are the tier rates and the season record.
went on to break out
improvement candidate tier
62.8% is the historical rate for this tier, not a probability for this player. The two tiers are shown together because a rate without its comparator says nothing.
wOBA lift against comparable players at the same production level — not against the league. Three held-out constructions, each significant at 90%.
Percentage-point lift in breakout rate over a matched control. Every CI 2022–2025 excludes zero; 83 out-of-sample weeks contribute.
2026 · watch. +8.9pp with a CI crossing zero — not distinguishable from no effect, and never run as a headline.
- 80% interval coverage
- 80% / 81% vs 80% nominal
- MAE, OPS points
- 131 30d · 101 60d
- Bias
- −1p / −3p
- Calls published
- 4,439 all public
since 2022
Every one public and sortable by the date it was called — the calls that worked and the calls that did not, in the same list, at the same weight.
Open the ledger →Board readings render live and change nightly; the measured, published figures on this page are the tier rates and the season record.
Everything we predict, we prove.
Four domains. Each one predicts and each one is scored — including the one that currently publishes nothing.
Player projections
Held-out accuracy for projections has not been published. The slot exists because the pairing rule does not bend. No figure appears here until one is measured.
Daily calls
RebuildingThe engine behind the old record was defective and has been replaced. A full-season walk-forward is running now. Figures appear only if that record clears its pre-registered gates; if it does not, nothing is published at all.
Superseded, not retiredBreakout & regression signals
Every flagged player, and what his surface line did in the weeks after.
Rest of season, week by week
Week-by-week scoring has not been published. Slot and route exist; the figure appears when the weeks have been scored.
What we do not claim.
On the home page, at the same size as the results. If one of these changes, it changes here first.
Every number here carries its comparator.
A lift figure is meaningless without saying over what. Ours is measured against comparable players at the same production level — the hardest available comparison, and the one that makes the number smaller. A tier rate is shown beside the weaker tier’s rate. An interval is published with its coverage. The methodology is ours; the result is public.