A little foresight. A better morning.
Tomorrow, made ready.
A considered bake sheet, from the rhythms of your neighborhood.
Monday, November 9, 2026
12°C · dry skies · regular trading
The bake sheet
10 daily favoritesRecommended bake—
Expected demand—
Production cost—
Preparing 18 months of sales history and fitting your bake forecast…
| From the oven | Forecast | 80% range | Bake |
|---|
Forecast = expected units sold · Bake balances missed sales against day-old value.
Behind the forecast
Actual salesModel demandSold out · lower bound8-week holdout
How the model works
Each product has its own log-space least-squares fit: weekday, month, temperature, rain, snow, holiday, local event and promotion. Sold-out training days are excluded because observed sales are only a lower bound on demand. The final 56 days are never used to fit. The bake target is a lognormal newsvendor quantile: (price − cost) / (price − salvage). The range is the 10th–90th percentile of demand, based on training residuals.
A little accountability
Holdout MAPE ↓| Product | Model | Last week |
|---|
Comparing unseen days with the same weekday a week earlier.