Evaluate forecast quality with transparent metrics and an honest chronological baseline.

This offline Python toolkit joins predictions to actual outcomes at matching timestamps and reports MAE, RMSE, sMAPE, bias, pinball loss, quantile calibration, interval coverage and width.

Included • Six-page PDF evaluation guide • CSV validator and exact-timestamp metric evaluator • Rolling one-step baseline that uses only preceding observations • Synthetic sample data, predictions and example report • Tests for known metrics, crossing quantiles and outcome leakage • Editable experiment register and evaluation report template

Uses P01, P25, P50, P75, P90 and P99. Python 3.10+; standard library only. No API account is required.

Instant ZIP download. Measures forecast error and calibration; it does not simulate trading returns. The sample is synthetic. Use and adapt for your own projects and client deliverables.