The comparison window decides the answer: which do you report?

A traffic drop measured over twenty eight days against the previous twenty eight days is a different claim than the same drop measured against the same period a year earlier. I have watched both numbers presented in the same meeting and produce opposite decisions about the same site.

The previous period comparison catches recent change, and it reports ordinary seasonality as if it were damage. The year over year comparison survives seasonality, and it hides anything that started six weeks ago. Neither one is wrong. Reporting only one of them, without saying which, is.

What I do is state the window before the number, every time, and show both whenever the two disagree. When they disagree, the disagreement is the finding, because two windows that point different directions locate the change in time better than either number does on its own.

Where I still lose the argument is the first month after a launch or a migration. Year over year is comparing two different sites, and the previous period is comparing against a week nobody wants to treat as normal.

What do you fall back on in that first month?