Ongoing analytics
Forecast accuracy and bias
Redbud Fabrication · monthly forecast review · through June 2026
Forecast against actual
Bias by segment
Mean signed error. Right of zero is over-forecast.
Is the bias closing?
Rolling six-month bias — aggregate against the two worst segments
Segment detail
MAPE measures how far off. Bias measures which way. When they match, the error is entirely directional.
Why this report exists
Forecast bias returns quietly. It does not announce itself in a monthly revenue review, because over-forecast and under-forecast segments cancel on their way up to the summary — that is exactly how it went unnoticed here for two years.
So the one thing this report does that a standard revenue pack does not is refuse to show the aggregate on its own. Every segment's signed error is visible to the people producing it, every month, which is the entire mechanism by which the number improves.
Switch to the last twelve months and watch what happens. Aggregate bias halves, from +3.7% to +1.9% — which reads as real progress. Gross segment error moves from 11.3% to 10.3%. Almost nothing improved; the errors just happened to cancel more neatly over the shorter window. That is the whole concealment, reproducible in one click, and it is why this report leads with the segment view rather than the summary.