Azadpur has no Medium/Small actuals on file — only Azadpur Cherry Large could be scored.
Actuals date format differs by market (Azadpur is DD-MM-YYYY; all other markets are MM/DD/YYYY) — parsed per-market.
A few actuals rows carry mislabeled or whitespace-inconsistent Grade values (e.g. Narwal Small, Ganderbal Medium) — matched by the file's own grade, not the in-row label.
Only dates with both a forecast and a matching actual are compared — forecast dates with no matching actual (non-trading days, or dates beyond the latest actuals on file) are excluded rather than scored against a missing value.
How to read this report
Price Accuracy
How close forecasts were to actual prices on average, as a single %, easiest to scan (100% − average price error). 80%+ is a solid forecast; below 65% means the market/grade needs attention.
MAPE (Mean Absolute % Error)
The average size of the forecast error, as a % of the actual price. Price Accuracy above is just 100 − MAPE.
RMSE / MAE
The typical forecast error in rupees per kg (RMSE weights large misses more heavily than MAE).
Directional accuracy
% of days the forecast correctly called whether the price would go up or down from the day before — useful even when the exact price is off.
Interval coverage
% of days the actual price landed inside the forecast's own lower/upper band. Close to 100% means the model's uncertainty range is trustworthy; well below it means the band is too narrow.
Overall accuracy
Model accuracy by market & grade
A single Price Accuracy score per market/grade — hover a tile for the underlying numbers.
Full metrics detail
MAPE by market, grouped by grade
Directional accuracy by market, grouped by grade
% of days the forecast correctly called the direction (up/down) of the next day's price move.
Error drift over forecast horizon
Mean absolute % error, bucketed by days since each series' forecast start (pooled across markets, split by grade). These are recursive multi-step forecasts, so error is expected to grow with horizon.