All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=791122

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=c5cda4d40dcd9b1d03bde8f4c7c758653c9e6248d6946379ecf91890812eb6b7


Bot(s) for this bug's original alert(s):

webrtc-android-tests-nexus5-kitkat

I looked back in the graphs, and there were ~5 alerts the past 5 months. Out of those, 3 were legit. I think that's an acceptable S/N ratio.

Out of those 3, how many were only detected on the nexus 5? Moreover, if there is no cause and no action to be taken then maybe sheriffs have been ignoring alerts instead of filing bugs?

Fwiw, I have had at least 3 alerts involving the nexus 5 during the last week, all of them false.

Note that it isn't noise in the normal sense of a drawing random values from a distribution. Here there are extremely clear and persistent changes in the metric without any apparent cause. It would be good to at least understand what affects the tests in this way.