A model, not a take
Every ranking on MatchWiz comes straight from a statistical model — there are no manual picks and no hot takes. The same math scores every hitter and every starting pitcher on the slate the same way, so a star and a backup are judged on identical terms: the matchup in front of them, not their name.
What goes into the score
Every score starts from the matchup in front of the player. For a hitter, that means weighing how he's actually performed against the kind of pitcher he's facing — handedness matters enormously, so a left-handed bat's history against righties isn't treated the same as his history against lefties — set against what that pitcher, and the bullpen behind him, tend to give up. From there the model accounts for how much the hitter is likely to play, the ballpark, the day's weather, and how hot or cold he's been lately, leaning on what's recent over what's stale. Pitcher boards work the same way in reverse — how the starter projects against the specific lineup he's drawing that day. The result is one comparable score for every player on the slate.
Those are the ingredients, and they're no secret — most serious models look at the same kinds of inputs. The edge is in how they're weighted and combined, and that part is our own: it's tuned and re-tuned against real outcomes every week. So rather than publish the recipe, we'd rather prove the model works the only way that actually counts — by grading every call in public, where you can check it.
What the score means — and what it doesn't
Every player carries two numbers, and they answer different questions.
The score runs 0–100 and rates the matchup, not the player. It's relative to the rest of the slate: a 90 means this is one of the better spots on the board today, not that something is 90% likely to happen. A great hitter drawing an ace in a pitcher's park on a cold night can score below an average hitter facing a struggling arm in Coors. That's the model working, not the model broken.
Confidence is coverage, not certainty. It tells you how much real data sits behind the score — plate appearances in the relevant splits, how much of the opposing arm's profile we can see, whether the lineup is confirmed. A rookie with 11 career plate appearances against lefties gets a low confidence number no matter how good his score looks. Read them together: a high score with low confidence is a thin read, and we'd rather show you that than hide it.
Neither number is a win probability. If you want one of those, the moneyline board publishes actual game-level win probabilities beside the book's implied odds — that's a different model output, and it's labeled as such.
Projections meet the market
A model number only means something next to a price. So wherever a book posts a line, we show our projection beside it and flag where the two diverge. The largest, best-supported gaps become Plays of the Day — the handful of spots each slate where the model most disagrees with the market. They're a research signal, published with their full graded record, not betting advice.
The market is the hard part, and we treat it with respect. A posted line already contains most of what a good model knows — the books employ sharp people and move fast on lineup news. So a gap between our number and theirs is usually the market being right and us being wrong. The gaps worth anything are the ones that survive: the same kind of disagreement, in the same kind of spot, grading out ahead over hundreds of settled plays. That's a much higher bar than "our number is different," and most divergences never clear it.
Where the numbers come from
Schedules, lineups, and box-score results come from MLB's official feed — the same source that settles the games. Book lines come from sportsbook odds feeds, refreshed through the day, with extra pulls close to first pitch so what we grade against is near the closing number rather than a stale morning price. DraftKings salaries feed the fantasy and DFS boards.
Boards rebuild each morning, then update through the day as lineups get confirmed and prices move. During games, results update live. A player who was projected but doesn't make the official lineup gets marked as scratched and drops to the bottom of the board with his rank removed — he's not quietly deleted, because pretending we never ranked him would be the dishonest version.
How grading actually works
A projection is graded against the real box-score result, at the line the market offered when we called it. Three rules keep that honest:
- The score is locked at first pitch. Once a game starts, its numbers can't be rewritten. No retroactive sharpening.
- The price is frozen with it. A play is graded at the odds it was posted at, not at whatever the closing number happened to be. If a line moves our way after we call it, we don't get credit for the better price.
- Once a play surfaces, it stays. It keeps showing and keeps getting graded even if the live edge drifts away before first pitch. Quietly dropping the ones that stopped looking good is the oldest trick in this space.
The full archive is public and rolls back date by date. Pick any board, walk it backwards, and read exactly what we said and what happened — including the days it went badly.
What the model is bad at
Most sites in this space don't publish this section. It's the most useful one.
Efficient markets beat us. The moneyline is the sharpest, most liquid market in baseball, and our game-level model grades close to break-even against it. Game totals are much the same. We publish those numbers anyway, as context and as transparency — but we don't dress them up as plays, because the honest read on a well-priced market is that there's nothing there.
Several prop markets look efficient too. We've run full reviews on home runs, total bases, and hits+runs+RBIs and found no edge we could exploit — the books price them about as well as we can. Those boards stay up because the projections are genuinely useful for lineup and DFS decisions, and because burying a board the moment it stops producing plays would leave you with a misleadingly rosy picture of what the model can do.
First-inning markets are hard. Roughly half of MLB first innings go scoreless and the books know it. We publish NRFI probabilities and fair odds as transparency, not as a play type.
Where the model has found real room is the softer, less-efficient corners — team totals more than game totals, and the strikeout market, where projected workload and the specific lineup a starter draws matter more than a season-long rate. Those are the boards that surface plays. The rest earn their place by being accurate, not by being profitable.
Known limits
- Not every player has a line. Books don't price every prop for every player. Even for confirmed starters, coverage tops out around 87% on a typical slate — so "no line" usually means no book posted one, not that we failed to find it.
- Late scratches happen after we publish. Lineups change minutes before first pitch. We mark scratches as soon as the official lineup lands, but a board loaded at 4pm can go stale by 7.
- Small samples stay labeled as small samples. Early in a season, after a call-up, or in a narrow platoon split, there simply isn't much data. That shows up as low confidence rather than as false precision.
- Pitchers returning from injury are the model's worst spot. A starter coming off the IL is often on a strict pitch limit that no stat line reveals. We flag long layoffs, but a projection built on his healthy workload will overshoot a 65-pitch rehab start.
- Any single day is noise. The model is tuned on hundreds of graded outcomes. Judging it on one slate — good or bad — tells you nothing.
We grade ourselves, in public
Every projection is graded against the real box-score result, at the line the market actually offered. The score for a game is locked at first pitch — once a game starts, its numbers can never be quietly rewritten to look smarter after the fact. And the entire archive is public: roll back to any past date on any board and read exactly what we said and what happened.
Season track record
As of the current season, MatchWiz's highest-conviction calls — the Plays of the Day, where the model's projection diverges most sharply from the market line — have hit at 58% over 166 graded plays, a +5% return on a flat one-unit stake. That edge concentrates at the top of every board: the highest-ranked spots clear the field by a clear margin, while the full board — every matchup we score, most of them ordinary — sits near the rate you'd expect against the vig. We publish all of it. Each projection is graded at the line the market actually offered, the score for every game is locked at first pitch, and the complete archive rolls back day by day so anyone can audit exactly what we said and what happened. The record is the product — not a marketing number, but the same ledger that tunes the model.
It learns every week
Each week the model's weights re-tune against what really happened, scored on a held-out test split so it can't just memorize the past. It's a continuous learning loop run out in the open: the model gets sharper as the season's graded sample grows, and the record above moves with it.
Re-tuning has a rule attached: a change has to prove itself on data it hasn't seen, across a full sweep of weightings, graded in units — not picked because it looked better on the days we already knew about. Plenty of proposed changes fail that test and get thrown out. A stat that stops earning its weight loses it, even if it's one we like.
How to read a board
Every board is the same slate sorted by one question. Open any of them and you get a ranked list of today's matchups, each row showing the score, the confidence behind it, our projection, and the book's line where one exists.
Tap any row and it opens. Inside is the case for the number: how the player has performed in this split over the last two weeks, the season, and two years; what the opposing starter gives up to that handedness; the bullpen behind him; park and weather; his last five games. That's deliberately the whole argument — if you disagree with the score, you can find the specific input you disagree with rather than arguing with a black box.
A few conventions worth knowing. Ranks reflect the current sort, so they shift as lineups confirm. A dash instead of a rank means the player was scratched. Boards carry their own graded record right on the page, and the date navigation walks the archive backwards. Unfamiliar terms are defined in the glossary.
How to check us
Don't take the record on this page at face value — it's our page. Audit it instead.
Pick a board, roll it back two or three weeks, and read the days one at a time. Every dated page shows what we published that morning and how it graded that night, at the line we called it at. If the top of the board didn't beat the bottom, that's visible. If a play type went cold for a fortnight, that's visible too. Compare a few of our projections against where the market closed. The archive is the argument — everything else on this page is just us describing it.
Who's behind it and how to reach us is on the about page, and contact goes to a real inbox. If you find a number here that's wrong, tell us and we'll fix it.
See it on today's board
The methodology is only as good as the board it produces. Start with today's matchups, check the day's Plays of the Day, or see the team-level model on moneyline & run totals. Every board carries its own graded, roll-back-able record.
How MatchWiz works — FAQ
What is MatchWiz?
MatchWiz is a daily MLB matchup model. It scores every hitter and starting pitcher on the slate from the underlying matchup — handedness splits, the opposing arm and bullpen, ballpark, weather, and projected plate appearances — then posts each projection next to the market line and grades itself against the real box-score result.
How does the matchup score work?
It weighs how a player has performed against the kind of opponent he's facing — handedness matters a lot — against what that opposing pitcher and bullpen tend to allow, then accounts for expected playing time, the ballpark, the weather, and recent form. Every player is scored by the same model, with no manual picks. The exact way those factors are weighted and combined is our own, and it's tuned every week against real results.
Are the projections or Plays of the Day betting advice?
No. They're a research signal from a statistical model, published with the full graded record so you can judge them yourself. The model is continuously tuned and can be wrong on any given day. Nothing here is a guarantee or a recommendation to wager.
How is the track record graded?
Every projection is graded against the actual result at the line the market offered, and the score for a game is locked at first pitch so nothing can be quietly rewritten after the fact. The entire archive — every past slate and how it landed — is public and rolls back date by date.
How often does the model update?
Boards rebuild each morning and refresh through the day as lineups and lines move; live results update during games. Each week the model's weights re-tune against what actually happened on a held-out test split, so it sharpens over the season.
What does the 0–100 matchup score actually mean?
It rates the matchup, not the player, relative to the rest of that day's slate. A 90 means it's one of the better spots on the board today — not that there's a 90% chance of anything. A star hitter facing an ace in a cold pitcher's park can score below an average hitter facing a struggling arm in a hitter's park. That's the model working as intended.
What is the confidence number?
Confidence is coverage, not certainty. It shows how much real data sits behind the score — plate appearances in the relevant split, how much of the opposing pitcher's profile we can see, whether the lineup is confirmed. A rookie with 11 career plate appearances against lefties gets low confidence no matter how strong his score looks. A high score with low confidence is a thin read, and we show you that rather than hiding it.
Why does a player show 'No line'?
Because no sportsbook posted one for that prop. Book coverage tops out around 87% even for confirmed starters on a typical slate, and lines for some players never appear at all. It's not a gap in our data — we show the projection either way, and grade it only where a real market price existed.
What is MatchWiz bad at?
Efficient markets. Our game-level model grades close to break-even against the moneyline and game totals, because those markets are sharp and already price in almost everything. Full reviews of the home run, total bases, and hits+runs+RBIs prop markets found no edge we could exploit either. Where the model has found real room is softer corners — team totals and the strikeout market. We publish the efficient boards anyway, as projections and as transparency, rather than hiding the ones that don't produce plays.
Can I check the track record myself?
Yes, and you should. Every board has a date navigation that rolls the archive back day by day. Each dated page shows what we published that morning and how it graded that night, at the line we called it at — losses included. The score for each game is locked at first pitch and the price is frozen with it, so nothing can be rewritten after the fact.
Do you delete plays that go badly?
No. Once a play surfaces it stays shown and stays graded, even if the edge drifts before first pitch and even if it loses. Both the board and the day's win-loss ledger are built from the same set, so they can never disagree.
Is MatchWiz a sportsbook?
No. We take no bets, hold no money, and settle nothing. MatchWiz is a sports data and projections service that grades its own output in public. Nothing here is betting advice or a guaranteed outcome. 21+, and please play responsibly (1-800-GAMBLER).