Proof · Finding
A number that changes with the window is not a number
Same data. Same day. Same dashboard.
over 60 days
over 90 days
over 120 days
Nothing changed except how far back the window reached. On seasonal revenue, a trailing comparison does not measure performance, it measures where you happened to put the edges.
This is the defect class that makes a tool confidently wrong. Not broken, not obviously unreliable: confident, specific, and pointing the wrong way. A weekly report built on a trailing window will tell a founder their business is growing and shrinking in the same month, and both readings will look rigorous.
How to tell if yours has it
Run the same comparison at three window lengths. If the sign flips, the metric is a function of the window, not of the business. Fix the window before you act on the number.
Source: our own analysis, 2026. No client identified.
Also in proof: what we got wrong, our own audit of every claim we had in circulation.