The gap that starts it
The line between a projection and a Play of the Day is 1.5 strikeouts. When our number for a starting pitcher runs at least that far above or below the book's line, the play posts. When the gap is smaller, the projection stays on the board as information — visible, honest, with no lean attached.
A strikeout prop works simply. A sportsbook posts a number before the game — say, 5.5 Ks for a starter — and you bet whether he goes over or under. We run our own number from the same matchup: the opposing lineup's contact rate, the starter's recent stuff, the ballpark and the umpire's zone. When our projection is 7.2 and the line is 5.5, the gap is 1.7. It clears. The play goes up. When the projection is 6.8 against the same line, the 1.3 gap doesn't clear, and it stays as a number you can look at and judge yourself.
That 1.5 threshold came from backtesting. Two full seasons of gaps were measured against real box-score results — how often a gap of at least 1.5 actually produced a winning bet, at what price, across every starter in the sample. At 1.5 and above, the win rate held at 62% or better, with positive returns somewhere in the 11-to-25-percent range depending on how you slice it. Below 1.5, the relationship gets murky. The threshold is where the signal held up under a real test.
Team totals work the same way, at the game level: how many runs we think one team will score versus the book's number for that team. When the disagreement clears a threshold there, that surfaces too.
Before it posts
Two gates stand between a qualifying gap and a posted play.
The first is odds. Our number only means something set against a real price. If a sportsbook hasn't posted a line yet — roughly one in eight confirmed starters never gets one on a typical slate — there's nothing to attach a play to. The projection sits on the board regardless. You can see our number and how far it is from the market. But without a price to freeze it at, no play posts.
The second gate is the lineup. A projection built on a player who gets scratched two hours before first pitch is a projection for someone who isn't playing. We don't post until the relevant lineup is final — the batting order confirmed, the starting pitcher locked in. A gap built on an unconfirmed roster is a gap built on sand.
When both are clear — odds in, lineup confirmed, gap above the threshold — the play posts.
The freeze
The moment a play posts, two things lock: the line and the price.
The line lock matters because markets move. A book might open a starter at 5.5 Ks and shade to 6.0 by game time as early action comes in. If our play posted at 5.5, that's the number it grades against. Whatever the book ended up at is the book's business. The play was called at 5.5.
The price lock matters more than it sounds. Odds shift constantly between when a gap is detected and when first pitch arrives. A price of −115 in the morning might be −130 by evening — that's a real difference in returns over a full season. We freeze the price the book was offering when the play surfaced. Win or lose, that's the price it grades at.
There's an honesty reason beyond the accounting. A system that grades against the final closing price — wherever the market settled — is taking credit for movement it didn't predict. It's grading a bet it never actually made. The frozen snapshot is the call we made; the market's final number is the market's subsequent opinion. Using one to grade the other is quiet score-padding. The play posts at a real number, at a real price, and that's where it settles.
No take-backs
Once it's up, it stays up.
The card stays on the board through game day with the frozen line and price visible. It grades against the real result at that frozen number. It lands in the day's win-loss count whether or not the gap that qualified it still exists by first pitch. The market might have moved the line from 5.5 to 7.0 after the play posted. Doesn't matter. The call was made at 5.5.
Nothing comes off. Not if the market moved further against the position. Not if the matchup projection shifted as lineups settled. Not if it goes 0-for-5 that week.
The archive rolls back date by date. Pick any past day on the plays board and you'll see exactly what posted that morning — the line, the price, the projected number — and exactly how it graded that night. All of it. Every play in that history is graded at the number it was called at. Wins and losses together.
This isn't stubbornness. A record that only shows plays that looked good in hindsight, or quietly replaces a bad number with a better one after the fact, doesn't tell you anything about the model. It tells you the publisher knows which ones to remember. The plays on the board and the day's win-loss count are built from the same set, so they can never say different things about the same day. If a play is on the board, it's in the count.
The same ledger that tunes the model
The record isn't just for you.
Each week, after results settle, the weights in the matchup model re-tune against a held-out set of real game outcomes — actual box scores, not what any sportsbook thought the number would be. The season record on the plays board feeds that process directly. When the model is running cold on strikeout projections, the next weekly tune sees it. When a particular market is finding real edges, that shows up too.
The record is public before the model reads it — not the other way around. No play qualifies because it would have graded well. The model produces the gaps; the gaps either clear the threshold or they don't; the plays post; they grade; the model tunes. Forward only.
That's what makes a cold stretch honest rather than uncomfortable. Three weeks of 3-for-11 means three weeks of signal for the next tune. A run like that might mean the market adjusted to something the model hasn't caught up with yet. Hiding it would be hiding the information the model needs. The misses are how it learns.
What this doesn't say
It doesn't say plays are common. On a typical slate, there's one, maybe two. On some days, none. When no starting pitcher's gap clears 1.5 against the posted lines, the board is empty. That's not a malfunction — it means the model and the market agree closely enough that there's nothing worth posting. Filling an empty board by lowering the threshold would mean posting plays that don't have a track record. That's the thing this whole setup exists to prevent.
It doesn't say the play will win. 62% is the backtested rate on strikeout-prop gaps at 1.5 and above — which means four of every ten plays have historically been losses. There are weeks where the board goes 3-11 and nothing seems right. That's baseball. A prop is close to a coin flip more often than anyone wants to admit. The edge comes from consistently finding spots where the price is a little off, over a long enough sample. One game tells you nothing. A season tells you something.
And it doesn't say every market has a play. Most don't. Home runs, hits, total bases, the game-level moneyline — full seasons of testing found nothing worth posting in any of them. The books price those well, and we haven't found an angle they're missing. Those boards are on the site for the projections and for transparency, but without a lean attached. Two honest edges matter more than a shelf full of manufactured ones.
Sources
Every number here comes from the graded ledger the boards publish. Research from a statistical model, not betting advice; no outcome is guaranteed. Method: how MatchWiz works.