How the Model Works
No black box, no cherry-picked wins. This page explains exactly where the numbers come from, how often they update, and how every single prediction gets graded in public — including the ones that miss.
The daily cycle
Data refresh
Schedules, probable pitchers, player game logs, platoon splits, and park factors are pulled from MLB's official league data feed. This runs continuously through the day — the model never works from yesterday's snapshot.
Projections form
Every batter expected in a lineup gets home run, hit, strikeout, and walk probabilities built from recent performance, the opposing pitcher's profile, handedness matchups, and the ballpark. Until official lineups post (usually mid-afternoon ET), lineups are projected from each team's recent starts against same-handed pitching — and clearly labeled as projected.
Market lines join
Sportsbook odds from major U.S. books are matched to each batter and game and refreshed throughout the day, so you can see where the model and the market agree — and where they don't.
Freeze at first pitch
When a game starts, its last pre-game projection is frozen. That exact snapshot is what the site keeps displaying for the started game, and it is the only thing grading is allowed to score. Predictions are never edited after the fact.
Grade when final
Once a game goes final, the official box score is ingested and every frozen prediction is scored against what actually happened. Only batters who actually took a plate appearance are graded; postponed and suspended games are graded whenever they finish.
Publish the results
Hit rates, calibration, Brier scores, and model-versus-market comparisons are published on the public Accuracy page — generated from the same graded records, not a hand-picked highlight reel.
What we predict
Batter props. For every batter in today's lineups: the probability of a home run, a hit, a strikeout, and a walk — shown both as percentages and as American odds so they're easy to compare against a sportsbook line.
Game predictions. Each matchup is simulated thousands of times using the actual (or projected) lineups, starting pitchers, and bullpens, producing a win probability and expected run totals for both sides.
Daily boards. The same numbers power the slate views — today's top home run candidates, matchup scores, and lineup-by-lineup projections.
Where the data comes from
Everything starts from MLB's official league data feed: live scores, lineups, game logs, splits, probable pitchers, park data, and pitch-level detail. Sportsbook lines come from an odds feed covering major U.S. books. Our own graded prediction history — every frozen snapshot and its outcome — lives in our database and feeds the accuracy reporting.
There are no paid picks, no insider whispers, and no manual overrides of the model's output. If the data doesn't support a number, we'd rather show nothing than make one up.
How we grade ourselves
The core rule: predictions are frozen before first pitch and never touched again. Grading only ever scores that frozen snapshot, so there is no way to quietly rewrite a bad call after the game. What you saw before the game is what gets graded.
After each game goes final, official box scores are ingested and every frozen prediction is scored. The Accuracy page reports hit rates by probability tier, calibration (when we say 30%, does it happen about 30% of the time?), Brier scores, and how the model performed against the sportsbook consensus — all computed from the same graded records, misses included.
Model versions & tuning
Every prediction is tagged with the model version that produced it, and accuracy numbers are never blended across versions — a new version starts its record from zero instead of inheriting (or polluting) the old one's track record.
Once a week, an automated backtest replays recent history using only the information that was available before each game — no hindsight allowed — and compares candidate adjustments against the current model on the exact same games. Changes are rejected outright when the sample is too thin to be meaningful, and every run stores a before/after report.
What the model can't do
Baseball is ruled by variance. A 6% home run probability is a real edge worth knowing about — and it still means no home run about 16 times out of 17. Single-game outcomes prove nothing in either direction; the calibration record over hundreds of games is the honest measure, which is why we publish it.
The model only knows what's in the data. Late scratches, sudden weather, a quietly injured hitter, or a bullpen decision made mid-game can invalidate a pre-game number. Early in the season, samples are thin and projections are less stable. Market lines also keep moving after we snapshot them.
And plainly: no model guarantees profit. Treat these numbers as one well-documented input to your own research, not an oracle.
Walk through one real prop — from today's lineup to the graded result, every step on live tools →
Responsible use
FullCountProps is for entertainment and informational purposes only. It is not a sportsbook, does not accept wagers, and nothing on this page or anywhere else on the site is betting advice. If you choose to bet, do so legally and responsibly — see our Responsible Gaming page. 21+. 1-800-GAMBLER.
Questions about the methodology? Ask us directly — every submission is read.