There was a typo that made the threshold 100x too low (0.16% instead of 16%). Should be fixed today.

Another change collided and made the metric sources change before this landed. Waiting on another change to check that it's fixed.

Tentatively declaring this fixed. Changes are up, and the alert is tracking the metric.

Declaration is tentative because the Pcon graph for the alert, http://shortn/_EnNA3BNB4a, and the corresponding viceroy Omen, vi/chromeos/provision#_VG_huYBJmlb, don't seem to be tightly coupled.

Re #7: I think you're looking at the wrong graph - here it is side by side with the Omen's underlying precomputation: http://shortn/_wY1i7w2w69

Oh, excellent. Yes, those match up, that looks like the differences are entirely about how they're doing averaging.

I'm not sure how to match up the queries more precisely. That's desirable but, I think, a low priority.

Alert fired correctly today.