Learnings
Results by device
Filter the evidence to one device and every pattern is pooled again from the tests that ran on it.
A change that helps on desktop can hurt on a phone, and an average over both can hide either. The device filter on the Learnings page pools the evidence again for one device: the tests aimed at it, and the tests that ran on every device.
It recomputes rather than hides. Each pattern is pooled again from that device’s tests, with its own estimate, range and evidence level, so a pattern that reads as mixed overall can turn out to be consistent on one device and absent on another.
Where tests of the same thing disagree in direction and the split lines up with device, the Resolve mixed opportunity says so, which turns a confusing result into a test worth running. The headline, the featured patterns and the opportunities always cover every device, so the summary at the top never changes under the reader.