In T317766 we compared the self-reported service metrics against webrequest logs and it leaves some big questions. For example, the sampled hour shows 41,000 successful snapshot requests in the webrequest log, but the service recorded 118,000 successful snapshots in statsd.

It's important to fix these graphs to understand whether there's something wrong with the request flow which causes extra internal processing, and so that we can build reliable monitoring graphs showing the endpoint error rate.