Docs

Results sheet

Write-back goes to WhichForecast_Results: fingerprint, ranked methods, forecast, intervals, and optional inventory helpers.

Glossary

MASE (scale-free error), ADI (average inter-demand interval), PI (prediction interval), FVA (forecast value added vs naive), Method DNA.

Developer sideload

Run the task pane locally over HTTPS and sideload manifest.xml. See the repository README.

Sister product

WhichDist ranks probability distributions with AICc. Different engine; same family.

Understanding the error metrics

  • MASE — mean absolute scaled error vs a naive (or seasonal naive) benchmark. Leads the ranking. Lower is better; below 1 beats the naive.
  • MAE / RMSE — absolute and squared-error scales in the series units. RMSE penalizes large misses more.
  • sMAPE — symmetric percentage error. Prefer this over MAPE when zeros exist.
  • MAPE — warned when zeros would make it explode or lie.
  • ME (bias) — e = y − ŷ. Positive ME means the forecast was below the actuals on average (under-forecast).
  • FVA — forecast value added vs naive: does the method beat the simple baseline on this series?