FOR THE CURIOUS
HOW THEENGINE WORKS.
The math, the data, the discipline. If you like a peek behind the curtain, here it is — plain-English, no fluff, receipts included.
WHAT HAPPENS EVERY NIGHT
From the schedule drop to the read on your screen.
PULL THE SLATE
Tonight's games, starting pitchers, umpire assignments, weather at first pitch. Comes in from SportsDataIO once the slate posts around 1pm ET.
WAIT FOR THE LINEUPS
Batting orders drop 60-90 min before first pitch. We wait — projections that assume yesterday's lineup are worthless. Nothing ships until the real 9 are in.
SIMULATE THE GAME
Every game gets played out thousands of times against a Markov base/out simulator. Each plate appearance is drawn from the batter's true-talent rate given the pitcher, the count, the runners, the park. This is the spine.
ADJUST FOR REALITY
Pitcher pitch-quality grades from Statcast, per-batter platoon splits, park run factors, weather effects on batted-ball carry, umpire strike-zone tendencies. Every input is real — nothing is 'league-average' once the lineup posts.
COMPARE TO THE MARKET
The engine's simulated probability gets compared to Kalshi and the sharp Pinnacle line. Where our number diverges from the market's, we mark it as a read.
SURFACE THE READ
The read with the biggest divergence and cleanest supporting signals is what you see. Free tier gets tonight's headline read. Paid tiers get the full slate + the reasoning stack.
THE LAYERS
Five things stacked on top of each other.
Every read is the output of these five layers working in sequence. If any one of them is broken, the read is wrong — so we track them independently.
EMPIRICAL BAYES SHRINKAGE
Raw batter/pitcher rates over-fit in small samples. We regress each rate toward its cohort mean using empirical Bayes — a hitter with 30 PA vs. LHP gets pulled hard toward league average; a hitter with 400 PA gets left alone.
MARKOV BASE/OUT SIMULATOR
The spine. Games play out as a state machine over (runners, outs) states. Every plate appearance transitions to a new state via the batter's outcome distribution given the pitcher. Validated against RE24 (run expectancy) — the classic sabermetric fingerprint.
PITCH-QUALITY + MATCHUP
Statcast pitch grades feed vulnerability flags (does this pitcher hang a slider? does this batter destroy fastballs?). Platoon splits, park factors, weather, umpire K-zone all layer on top with bounded multipliers so no single input can dominate.
BAYESIAN CALIBRATION
Model outputs get shrunk per-market against actual settled results from the last 14 days. If the raw model says 65% and history says the model has been running 8pp hot on that market, we down-shrink. Refit weekly — an engine that stops calibrating stops earning trust.
MARKET DIVERGENCE + EDGE FLOOR
A read only surfaces when the engine and the market disagree by more than a price-tier-specific minimum. A 30-cent contract needs less divergence to be interesting than a 10-cent lottery ticket — the floor scales with price.
DISCIPLINE
Why you can trust the reads.
EVERY READ POSTS LIVE
Reads land in the model log the moment they surface. Wins, losses, cold streaks, all timestamped. Nothing edited after the fact. See the model log →
WEEKLY CALIBRATION
Layer 4 gets re-fit from the last 14 days of settled results every Sunday. An engine drifting away from the market gets caught and shrunk before it costs you.
THE [PAPER] TAG
Any engine without a 2-week track record ships tagged [PAPER]. It means: signal isn't proven yet — treat as research, not conviction.
QUESTIONS PEOPLE ASK
Straight answers.
Do you guarantee wins?
No. This is analytics, not a magic 8-ball. We show you where our engine's estimate disagrees with the market. Sometimes the market is right and we lose. We post the losses same as the wins so you can see when.
Are these sports picks?
We're a sports analytics platform. Our engine surfaces reads — signals where our simulation output diverges from the market price. You decide whether to bet on any of them. We don't place bets on your behalf, we don't collect from your bookie, we don't have a stake in whether you bet at all.
Where do the odds you compare against come from?
Kalshi (the prediction market) and sportsbook APIs. We compare our engine's probability to the market's implied probability from the current price.
How do you make money?
Subscription only. Free tier gets one read per night. Paid tiers unlock the full slate and the reasoning stack. Zero affiliate deals with sportsbooks. Zero cut of user bets. We don't win when you lose.
What's the [PAPER] tag mean?
Any engine without a 2-week live track record ships with a [PAPER] tag on its reads. It means we're still watching it — the signal might be real or might be noise. Once the sample is big enough to draw a conclusion, the tag comes off or the engine gets retired.
Why MLB first?
Baseball is the most-modeled sport in existence — Statcast alone gives us pitch-by-pitch data no other league has publicly. If our engine can beat the market anywhere, this is the easiest place to prove it. Other leagues follow on the same architecture as their data feeds mature.
READY TO SEE ONE?
The homework's done. Free every night.
18+ · No credit card · The read drops 1 hr before first pitch.