Just so I understand the background, is this to improve our ability to track? For example, we don't want giant long queries with lots of special cases so we wouldn't want to put "-status:Duplicate" in our check for how many performance bugs were closed without fixing?

What about if a bug gets manually un-duped? This happens sometimes. For example with CLs that have broad performance impact, there are multiple bugs and sometimes the developers prefer to keep them separate and work on the blink_perf.layout regressions separately from the memory regressions, or whatever makes sense to them.

It was to make it easier, but I didn't think about that case. I'll have to hardcode the queries so we get consistent numbers milestone to milestone.