ONYX grades every hitter on tonight’s slate with nine weighted models and probabilities fit to real outcomes. This page explains what each number means, how to act on it, and — just as importantly — where the model is weak. Nothing here is a tip sheet. It’s a manual.
If you read nothing else, read this. Four sentences that cover 90% of using ONYX well:
1
The score is conviction (1–99). The percentage is the real chance. They are different things — a 78 is not a 78% chance.
2
Next to every percentage is a fair price. If the book pays longer than fair, bet it. Shorter, pass. That single habit is most of the edge.
3
Straight bets on strong plays beat parlays over a season. Parlays multiply mistakes as fast as they multiply payouts.
4
A 25% play misses three times in four by design. Judge the model over a month on Results, never over a night.
02The score — one number, 1 to 99
Every hitter gets a score per market: HR (home run), TB (2+ total bases), HIT (1+ hit), and ONYX — the composite (45% HR + 35% TB + 20% HIT). 50 is a league-average spot.
PRIME80+rare — measured at ~1.9x the field's HR rate, on 55 graded games
EDGE60-79a real edge — the working zone, ~1.4x the field
EVEN50-59league average — no push either way
COOL40-49below the field — the model is cool here
COLDunder 40about half the field's rate
Why EDGE spans twenty points.
It used to split at 70. Then the graded record showed hitters at 70–79 homered 19.4% of the time and hitters at 60–69 homered 17.0% — a 2.4-point gap on a sample of 1,872. That is inside the noise, so showing them as two different grades implied a precision we do not have. They were merged.
A hitter marked no read has too little tracked data — every input defaulted to average. Skip those; they are not sneaky value.
03What builds the grade — nine models
The HR score is a weighted blend. The weights are earned: each model was backtested against real outcomes before getting a share, and anything that failed validation was thrown out.
Power50%Barrel rate, HR-per-PA, exit velocity — the engine. Nothing matters more than how hard and how often a bat squares the ball.
Form12%The last 7-14 days of real results blended with season expected stats. Hot is real, but skill anchors it.
Matchup8%Platoon splits against tonight’s starter’s throwing hand.
Pitch Vuln7%How hittable the opposing starter is — barrels and xwOBA allowed, with recent form blended in.
Pitch Match7%The hitter’s damage against the exact pitch types this starter throws, usage-weighted.
Park6%The yard’s HR factor. Coors and Yankee Stadium are not Petco.
Weather4%Temperature and wind fused into an HR index. 95 degrees with the wind out is a different sport.
Lineup Spot3%Where he hits in the order — mostly a proxy for how many plate appearances he gets.
Bullpen Exp3%A short starter means more innings against exposed middle relief.
Where this model is weak — stated plainly.
Lineup Spot is the softest component. Measured against real outcomes, the leadoff slot produces the highest HR rate on the board (17.6%, and the most plate appearances at 4.50) — but the model ranks it fourth. The heuristic correlates 0.823 with reality, which is directionally right and shaped slightly wrong. It carries 3% weight, so the damage is small, and it is on the list to re-fit.
04Probability & fair price
The score is conviction. The calibrated probability is the actual chance: “22% to go deep” means hitters in that band homer about 22% of the time in graded history. Re-fit weekly — and only applied when the new fit still beats its own baseline, so a thin sample can never poison the live curve.
The fair price converts probability to American odds. That is your break-even line. Longer than fair is value; shorter is juice on our own number. Try it — drag the probability and type in what your book is showing:
FAIR PRICE+355
EV PER $100+$0.10
THINbarely above break-even; fine at small stakes, easy to passmodel 22% vs book 22% — +0.0 pts of edge
Verify all of this yourself on Results: every forecast graded against what actually happened, misses included.
05Reading the board in 30 seconds
1
Pick your market
The board opens on HR because that is the market it prices: sorted by HR, the percentage and the fair price beside each name are the same market you are reading. Ranking skill is statistically indistinguishable between HR and the ONYX composite, so coherence decides it. ONYX, HIT and TB are one tap away.
2
Read the brief
Expected HR across the slate tells you if it is a launch night or a grind. The hero is the single loudest spot; the cells beside it carry best odds and biggest riser.
3
Scan the score column
Rows are ranked by the active market. The colour ramp runs muted slate (weak) to bright ice (elite) — brightness always means strength.
4
Check PWR and FORM
Two micro-bars per row. A big power number with a dead form bar is a different bet than both firing.
5
Tap for the full read
Expands to six component bars, the matchup line, and one tap to the full scouting page.
06The tools — one question each
The cheat sheet puts every market’s top plays on one page, so you rarely need the deep pages. Worth knowing: several tools are re-sorts of the same rows. These four genuinely surface different bats:
Weak ArmsRanks pitchers, not hitters — completely different list from the board.
Pitch EdgeHitter-vs-arsenal mismatches. Almost no overlap with the score board.
HotPure 7-day momentum. Surfaces bats the season-weighted board buries.
SplitsPlatoon-first ranking — the classic lefty-masher-vs-righty spot.
Multi-HR, Due and the Markets HR tab mostly re-rank the same top bats the board already shows. Useful framing, not new information.
07Straight bets — where you should start
One outcome, one bet. It is the highest-EV way to use the board and where serious bettors live. Read a row as: the market is the bet, the percentage is the chance, the odds beside it is break-even, and the big number is conviction.
HIT~60-70%Cashes most often. The grind that pays the bills.
TB~35-45%Middle ground — power without needing it to leave the yard.
HR~10-30%The swing. Big price, only when number and spot align.
Flat-stake the ones where the price beats fair, and let calibration work over a month rather than a night.
08Pairing legs — the parlay curriculum
Parlays multiply probabilities, which means they multiply mistakes. Most slips lose before they are placed, because the legs were paired wrong.
LEG 162%
LEG 258%
SLIP HITS36.0%
FAIR PAYOUT+178
1 IN2.8
Independent legs only — this is straight multiplication. Two bats in the same game against the same starter are not independent: one dominant pitching night kills both at once, so the true number is lower than this.
RULE 1 — KNOW THE REAL MATH.
Drag the sliders above. Two 60% legs are 36%, not “pretty likely”. Four coin-flippy legs are a lottery ticket with extra steps. The Parlay Panel on the board does this live for your actual slip.
RULE 2 — SAME-LINEUP LEGS HELP LESS THAN YOU THINK.
The folklore says two hitters facing the same starter live and die together. We measured it: across 482 games and 73,645 same-game pairs, two bats in the same lineup homer together barely more often than independence predicts — a correlation of 0.021 (95% 0.0085 to 0.0349), not the 0.2-plus the intuition implies. Two team-mates went deep in the same game 519 times against 521 expected. Real, and very small.
Hitters on opposite sides of one game measured independent — they share a park and a wind and nothing else. So a “same-game” slip built across both dugouts gets no boost at all, and the slip on the board no longer pretends otherwise.
RULE 3 — BUILD BY INTENT, NOT SCORE-SORTING.
“Take the top 3 scores” is lazy and usually correlated. SAFE takes the highest probabilities, one per game. VALUE pairs strong non-obvious grades books have not shaded. DART is raw-power longshots — one per slip, small stakes. A sound three-leg: two SAFE and one VALUE.
RULE 4 — MIX MARKETS TO CUT VARIANCE.
HR legs are 10–30% events; HIT legs are 55–70%. Three HR legs is a lottery ticket. HR + HIT + HIT keeps a real payout while multiplying a survivable win rate.
RULE 5 — CONFIRMED LINEUPS ONLY.
A projected bat who gets scratched is an automatic dead leg. Rows show IN or PROJ — check before lock, every time.
2-LEG (DEFAULT)SAFE + SAFE, different games, at least one HIT/TB. Target 30-40%.
3-LEG (STANDARD)2 SAFE + 1 VALUE, three games, max one HR leg. Target 15-25%.
STACK (CORRELATION)2 bats, same lineup: worth about +9% over the naive multiply on two 15% HR legs, nowhere near double. Measured r 0.021.
09The five most common mistakes
1
Reading the score as a percentage
A 78 is not a 78% chance. The score is conviction on a 1-99 curve; the percentage beside it is the actual rate. Always bet off the percentage and the price.
2
Betting a number without checking the price
A great spot at a terrible price is a bad bet. The whole point of the fair-price chip is that it turns a good player into a good wager only when the book cooperates.
3
Judging the model on one night
Calibration is a claim about hundreds of forecasts. A 25% play missing four times in a row is completely expected. Look at Results over weeks.
4
Stacking correlated legs by accident
Three bats off the same starter is one bet wearing a disguise. If that arm has a good night, all three die together.
5
Chasing the biggest number on the board
The top score is often the most heavily shaded price. The edge lives where the model is confident and the book is not paying attention — usually the 60-73 band.
10Discipline
The model is calibrated, not psychic. A 25% HR play misses three times out of four by design — the edge shows up over weeks. Flat-stake singles on EDGE+ plays with fair-price discipline beat any parlay strategy over a season. And when the slate is thin, no bet is the strongest bet.
Research and modelling output, not betting advice. Play within your means.
11Glossary
Every stat on the board, in plain English. Search it or filter by category.
41 of 41 terms
Barrel%Contact
Share of batted balls hit at the ideal exit-velo + launch-angle combo — the contact most likely to leave the yard. 8% is average, 14%+ is elite. This is the single biggest input to the Power model.
Exit velo (EV)Contact
How hard the ball comes off the bat, in mph. League average ~89; 92+ is serious thump. Averaged across all batted balls.
Max EVContact
A hitter’s hardest-hit ball of the season. A ceiling indicator — it says what the bat is capable of on its best swing, not what it does typically.
Hard-hit%Contact
Share of batted balls at 95+ mph. Broader and noisier than barrel rate, but stabilises faster in small samples.
xwOBAContact
Expected weighted on-base average — what a hitter’s contact quality says he should be producing, stripped of luck and defense. .320 average, .360+ strong.
xSLGContact
Expected slugging: the total-base version of xwOBA. Drives the Total Bases model at 57% weight.
ISOContact
Isolated power (SLG minus AVG) — raw extra-base pop with singles removed. .165 average, .250+ elite.
HR/PAContact
Home runs per plate appearance, the cleanest power rate. League ~3.2%. Regressed toward the mean for small samples so a 2-for-8 hot streak cannot fake elite power.
Pull%Contact
How often a hitter pulls the ball. Pulled fly balls are where most homers live, and pull-heavy lefties benefit disproportionately from short right-field porches.
Sprint speedContact
Feet per second on a competitive run. League ~27.0. Fast runners beat out infield hits, which lifts the HIT market more than the HR market.
Chase rate (O-swing%)Contact
How often a hitter swings at pitches outside the zone. Lower is better discipline. League ~28-32%.
HR/9Pitching
Home runs allowed per nine innings. ~1.2 average; 1.5+ is genuinely hittable. Blended 55% season / 45% last-14-days when a starter has 10+ recent innings.
xwOBA allowedPitching
Contact quality a pitcher gives up. The cleanest read on whether an arm is actually vulnerable or just unlucky.
Whiff%Pitching
Swing-and-miss rate on a given pitch. High whiff means the pitcher misses bats with it — a hitter who does not whiff on that pitch has a real edge.
ArsenalPitching
The mix of pitch types a starter throws and how often. ONYX crosses this against each hitter’s per-pitch damage to build the Pitch Match component.
xK / opp-adjusted xKPitching
Projected strikeouts for a starter tonight, scaled by how strikeout-prone the actual opposing lineup is.
IP/GSPitching
Innings per start. Under 5.0 means the bullpen gets exposed early — that is what the Bullpen Exposure model measures.
ONYX scoreModel
The composite: 45% HR + 35% TB + 20% HIT, on a 1-99 scale where 50 is a league-average spot. Good for a cross-market read; for a single market, use that market’s own score.
Calibrated probabilityModel
The % ONYX shows is fit to real graded outcomes: when it says 18%, hitters in that band go deep about 18% of the time. A measured rate, not a hunch. Re-fit weekly, and only when the new fit still beats its baseline.
Fair priceModel
The break-even American odds for a probability. 25% = +300. Beat that at a book and the bet has positive expected value.
Brier scoreModel
How good a set of probability forecasts is — lower is better. The baseline is "always guess the base rate". Beating it means the model adds real information.
Calibration gapModel
Predicted rate minus actual rate, in percentage points. Near zero means the numbers on the card mean what they say.
Projected vs confirmedModel
Before a lineup posts, ONYX scores off the team’s most recent lineup (PROJ). Once the real card drops it rescores (IN). A projected bat who gets scratched is a dead leg.
Regression to the meanModel
Small samples get pulled toward league average before scoring, so a hitter with 12 PA cannot post an elite grade off noise.
Poisson (2+ HR)Model
The math that turns a single-homer chance into a two-homer chance. Only elite power nights clear about 3%.
PRIMETiers
Score 80+. Rare — a handful per slate, 55 in the graded record so far. Those went deep 21.8% of the time against an 11.8% field: about 1.9x. Small sample, wide interval, and /results carries the live version.
EDGETiers
Score 60-79. The everyday working zone, about 1.4x the field. This band used to be split at 70, but the graded record showed 19.4% vs 17.0% either side — not a real distinction, so it was merged.
EVENTiers
Score 50-59. League average. No edge in either direction.
COOLTiers
Score 40-49. Below the field — the model is cool on the spot.
COLDTiers
Under 40. About half the field’s HR rate.
No readTiers
Too little tracked data — every input defaulted to average. Skip these; they are not hidden value.
HRMarkets
Hitter to homer. Low hit rate, big price. Only worth it when the number and the spot both line up.
TB (2+)Markets
Two or more total bases — a double, or two singles, or a homer. Power without needing the ball to leave the yard.
HIT (1+)Markets
One or more hits. Cashes most often (~60-70% on strong plays) — the grind that pays the bills.
F5Markets
First five innings. Starter-weighted: bullpen and lineup-slot factors drop out because you only face the starter.
H+R+RBIMarkets
Hits plus runs plus RBI combined. Lineup-slot dependent — run-scoring opportunity matters as much as the bat.
Park factorSlate
How much a yard helps or suppresses home runs vs neutral. 1.00 is neutral; Coors runs well above, Petco well below.
HR index (weather)Slate
Temperature, wind speed and wind direction fused into one number. 1.00 neutral. Hot air is thinner and carries; wind blowing out adds meaningfully to carry.
Wind-aid scoreSlate
Wind help specific to a hitter’s pull side, not just dead center — a pull-heavy lefty cares about the right-field line, and this scales by how pull-heavy he actually is.
xHR (game)Slate
Expected home runs in a game, summed from every hitter’s calibrated probability. 2.6+ is where stacks and overs live.
StackSlate
Multiple hitters from the same lineup. Deliberately correlated: if the starter is getting hit, they all benefit together.
ONYX model v1.0.0 · nine models · probabilities re-fit weekly against graded results · see the receipts