Name the figure and the direction first, while you still have a clean reading of where it stands. A recalibrated check should show up in Critical & Blocker or Pass Rate. Widening assignments should show up in coverage before it shows up in either. A change to how a team tests should show up in Change Failure Rate rather than Deployment Frequency.

Decide it in advance, because most changes move several figures a little and it is easy to pick the one that flatters the change afterwards.

Use the same time range and the same filters before and after. A saved view is the reliable way to do this, since it captures every active filter including the search term and brings the same scope back each cycle. See Filter activity by project, pipeline, database, or environment.

Every tile shows the current window followed by up to three earlier windows of the same length, newest first. A 30-day range reads Current, Previous 30 days, then 31–60 days ago. Hovering an entry shows its value and the exact dates it covers, and a window with too little data is de-emphasized rather than shown as a misleading zero.

Note: The comparison needs equal-length windows, so it is not shown at all when the range is Custom or All time. Choose 7, 30, or 90 days whenever you are checking whether something changed.

Deployment Frequency, Change Failure Rate, and Rollback Rate have a trend chart on the Database Changes dashboard, which is the better read when you want the shape of a change rather than two numbers. Cycle Time, the six policy figures, and the five drift figures have no chart, so for those the period comparison is the whole picture.

Note: The chart always plots the last five weeks, whichever time range you have selected, so it and the figure above it can describe different periods. Read the chart for the shape and the figure for the number.

One window after a change is rarely conclusive, because the window still contains work that started under the old rules. A change that has moved nothing after two full windows is usually telling you the constraint was somewhere else, which is a result worth having rather than a failure.