Continuous testing for descriptor parsing
Thanks to Karsten we have reasonably good assurance that we can parse past descriptors (#7828 (closed)). However, we should do some sort of continuous testing over old metrics archives to check for regressions.
This would require a bit of design (I'm not yet sure where this would live nor how it would work). That said, given the number of issues found on #7828 (closed) it could certainly be an effective method for finding issues. :P