THREE SPORTS · ONE STANDARD OF PROOF · onyx v1.0.0

Every player, graded.
Every miss, published.

Baseball runs tonight: all 270 bats scored for home runs, hits and total bases. Football and hockey are modelled the same way, on the same rule — every probability gets checked against what actually happened, and the misses are published at the same size as the hits. No locks, no touts. The board beside this is tonight’s real output.

Last 14 slates: the model said 12.6% to go deep. Hitters actually homered 11.5% of the time. How close those two numbers stay is the entire product.
TOP HR SPOTSLAST NIGHTWednesday, Aug 19 · 15 games
34.2EXPECTED HR
1
Conine MIAvs Nola (RHP)
27%+274
78
2
Ohtani LADvs Freeland (LHP)
27%+274
78
3
Schwarber PHIvs Alcantara (RHP)
25%+294
76
4
Alvarez HOUvs Ureña (RHP)
25%+305
75
5
Murakami CWSvs Holmes (RHP)
24%+315
74
6
Olson ATLvs Bradley (RHP)
23%+327
73
+ 264 more bats scored that night

Live output, not a mockup. % is the calibrated chance he goes deep; beside it is the break-even price you’d need to beat.

The numbers behind the numbers

Measured against graded outcomes, not asserted. Re-checked nightly on the track record.

0.592HR MODEL AUC

Ranking skill on real outcomes across 12,089 graded forecasts. 0.50 is a coin flip; anything durably above it is a genuine, repeatable read.

3.0×TOP VS BOTTOM DECILE

The best-graded tenth of the board homers at 19.2%. The worst tenth homers at 6.5%. That spread is the edge, stated plainly.

16,381GRADED OUTCOMES

Every prediction across 72 scored slates, joined back to the box score. Hits and misses both.

±1.2ptCALIBRATION GAP

When it says 20%, it happens about 20% of the time. Measured over the last 14 graded slates. Probabilities are re-fit weekly — and only applied when the new fit still beats its own baseline.

How a grade gets built

01

Ingest

Statcast contact quality, season and platoon splits, every starter’s pitch arsenal, live weather and park factors, and the confirmed lineup — pulled and rescored as cards drop.

02

Score

Nine weighted models led by Power at 50%: barrel rate, HR-per-PA, exit velocity. Then form, platoon matchup, how hittable the arm is, how the hitter handles that arm’s specific mix, park, weather, lineup slot, bullpen exposure.

03

Calibrate

The 1–99 score becomes a real probability through a curve fit to actual outcomes — so the percentage means what it says instead of being a vibe attached to a number.

04

Grade

Every completed game is pulled from the box score and joined back to what the model predicted. That record is public, including the days it was wrong.

What’s inside

The Board

Every hitter scored for HR, hits and total bases, with a fair break-even price on each. Grouped by game so a whole stack reads at once.

Player scouting

Barrel rate, exit velo, platoon splits, the arm he’s facing and how he handles that arm’s pitch mix — with percentile context on every number.

Weak Arms

The starters getting barreled, with their full arsenal broken out by usage, whiff rate and damage allowed per pitch type.

Pitch Edge

Slate-wide hitter-vs-arsenal mismatches — where a bat’s per-pitch damage lines up against exactly what he’ll see tonight.

Weather & Park

Live wind and temperature folded into every HR grade, all 30 park factors, and a wind compass oriented to the real ballpark.

The receipts

A public calibration table and full game log. Not a screenshot of one good night — the entire record.

ONE STANDARD, THREE SPORTS

Same rule everywhere: publish the misses

STRAIGHT UP

“I’m not selling you winners. I’m handing you the exact board I use — every number shown, every miss on the record — and letting you make the call.”

Research and modelling output, not betting advice. Play within your means.