How to use mlb·edge

mlb·edge predicts each game's home run margin, converts it to a win probability, then calibrates that against its own past results. It pulls in sportsbook odds and flags where the model and the market disagree. Two models run (runs and xwoba); the runs model plus a starting-pitcher adjustment drives the Today and Teams pages — it's the one that's validated best in walk-forward testing.

Everything is scored out-of-sample: predictions are stored before games start and graded once results arrive, so the Accuracy page is an honest ledger, not a backfitted one.

Games for today and the next two days, in collapsible per-day panels. Within each day, cards are sorted so the best-value bet floats to the top. Each card:

Field	What it means
Model line	Home team and predicted margin — "Tigers −2.56" = home favored by ~2.6 runs.
Win %	The model's calibrated chance each team wins.
Fair ML	The moneyline those odds imply — your break-even price.
ML (book)	The sportsbook moneyline (2-letter book tag, e.g. dr = DraftKings).
ML EV	Expected profit per 1u at the book's price. A green pill means the model rates it +EV. The small %K is the ¼-Kelly suggested stake.
Run line / Cover % / RL EV	Same read for the ±1.5 run line (favorite wins by 2+; underdog loses by ≤1 or wins).

Colored badges flag context: xwOBA form ↑/↓, running hot/cold (results vs expected), back-to-back, and rest edges.

The rule of thumb: when a book's price is worse than the model's Fair ML, EV turns positive and the game rises to the top. Thin or negative edges are the norm — that's expected.

Fastest way: on any Today card, open ＋ Log a bet. Each offered market (moneyline, run line) is pre-filled with the teams, side, book odds, and a Kelly-suggested stake — just adjust the units and hit Log.

Manually: the Bet Log page has a form — pick the date, Away and Home teams, the market (Moneyline or Run line), the Side (the team you're backing), a Line (±1.5 for run line), American Odds (e.g. −150), and a Stake in units.

Bets settle automatically when results come in, and roll into the P&L / ROI on the Accuracy page. Stakes are in units — treat it as paper-trading.

Every team ranked by Rating (model-expected run margin vs a league-average opponent, home/away blended). Plus the four model inputs (RF/g, RA/g home & away), last-10 record, xwOBA form (playing above/below season level), and Luck (actual run share minus expected — large positives tend to regress).

Models compared puts runs vs xwoba head-to-head: games scored, winner accuracy, Brier, log loss, run-margin error. Click a model name to load its detail. Below that: a rolling-20-game accuracy chart, a calibration table (does "58% predicted" actually win ~58%? — small gaps are good), and your bet P&L / ROI. This is the page to trust over time.

Lower Brier / log loss = better-calibrated probabilities; winner accuracy is a coarse tiebreak.

Every scored prediction, month by month: the pick, win %, predicted margin, actual result, and ✓/✗. Good for spot-checking how specific calls resolved.

The formula, the two-model explanation, and the fit history (coefficients over time). The Run pipeline now button forces an immediate sync → refit → odds → predict if you don't want to wait for the nightly 06:30 run.

• Trust the calibrated win %, not a gut read of the raw margin — the win probability is what's been validated (the model picks ~54% of winners vs a ~52% home baseline).
• Edges are thin against the ~4.5% book vig. Positive-EV flags are opportunities, not certainties.
• It updates itself nightly at 06:30 (results, refit, odds, predictions), so pages fill in and the track record grows on their own.