Override evidence

Forecast Value Added Auditor

Test whether released judgmental adjustments beat the statistical baseline after actual demand arrives.

Forecast outcomes

Enter 4–120 aligned periods. Use the baseline that existed before the human adjustment.

Method

For each period, absolute error is |forecast − actual|. Forecast value added (FVA) is baseline MAE minus final-forecast MAE, so positive FVA means the adjustment reduced error. Mean signed error shows whether the process tends to over- or under-forecast.

Why changed periods matter

All-period accuracy can dilute a harmful override process when most periods are untouched. This auditor reports both all periods and the subset where baseline and final forecasts differ, plus improved, worsened, and tied adjustments.

Limits

FVA measures historical error, not whether an override was reasonable when made. Sparse outcomes are noisy, costs can be asymmetric, and a small number of large events can dominate. Retain adjustment reason, author, time, information set, and evaluation window; group results by reason before changing governance.