A crash trend without a decision attached becomes a tour of charts. Before opening the reporter, name the choice: ship this candidate, hold for a fix, or rank backlog items for the next two sprints. The decision sets the window.
Match window to adoption
If only a thin slice of users is on the candidate build, a seven-day global view will drown new signals in older versions. Filter to the versions in play, then choose a window long enough for those versions to gather honest volume—sometimes forty-eight hours is enough after a wide rollout; sometimes you need a week on a staged release.
Keep a baseline close
Interpretation improves when the baseline is the last build you trusted, not an arbitrary quarter. Side-by-side windows expose concentrations that a single long slope hides.
Write the question on the brief
When you engage an external reader—or your own Friday afternoon self—put the decision and window in the first paragraph. Application analytics then serve the question instead of inviting endless exploration.