Assumptions:
- Swarming tasks have tags that helps "describe" the content of the task or at least, bucket them. For example, buildername, stepname, etc.
- Each dimension key value pair is automatically added as a tag, so these can be leveraged implicitly.
- There's a correlation overtime with a subset of the task tags and the task's duration.

Goals:
- Create a query engine that for a given list of tags, provide an estimated duration value within <500ms, preferably with confidence level.
- This can be used for multiple items: Determine cold/hot cache estimated durations, CQ SLO, throughput estimated, early load shedding, etc.


AIs:
https://xkcd.com/1838/