This metric is a more precise measurement of productivity loss than simply
measuring broken commits on the master branch; some jobs don't ever run on
PRs and thus will break master both without the author being able to predict so and without
other devs ever noticing.
Note: A few factors result in the most recent data being omitted:
Buckets start on a fixed day of the week (Sunday)
The current week is always omitted, as it is incomplete and would show misleadingly lower numbers
A master build failure span falls into a bucket based on the span's start timestamp
Some recent failure spans may not have been manually annotated yet
This query result is cached as a "materialized view" nightly (around 5:30am UTC)
Detected but not annotated
These are lateral and longitudinal clusters of failures that have not been marked by a human.
All Longitudinal Detected
These are commits for which one or more successive commits failed the same job.