Taskcluster has first class support for artifacts and treeherder also has the same for displaying them... We just need to listen to the artifact exchange and publish them to treeherder jobs as results come in (there may be races?)

Regarding races, taskcluster-treeherder has prefetch 1 to avoid any races: https://github.com/taskcluster/taskcluster-treeherder/blob/4f05ac58387d1a0df6d1789211e4d4d371502ae7/treeherder/handlers.js#L57 So we're hopefully not going to have race conditions, but we might eventually have performance issues. If this turns out to be an actual problem, we do some tricks like prefetching more and queue messages from the same taskId, so that different taskIds can be treated in parallel. We'll have race conditions if we try to scale taskcluster-treeherder beyond a single heroku node. At that point maybe we should just ask treeherder to consume our AMQP messages directly.