I think we had a bug open on this, but I can't seem to find it now. The "trigger-bot", implemented in bug 1163698, uses pulse to determine when a test job on try fails, and re-triggers it using the buildapi when it does. It keeps failure data for each push, and stops triggering after a certain percentage of the push has been re-triggered (I believe the threshold is 3% currently). In order to trigger only jobs that are visible by default, it determines which jobs are hidden by consulting Treeherder. Now that we're switching platforms to Taskcluster and turning off buildbot jobs, we should consider whether to implement this for Taskcluster. I have recently heard from developers that this feature is very useful when looking at a try push and attempting to determine whether they have introduced a failure.