We have that very big list of tasks, but this isn't very clear.
We can convert these to individually tracked issues - e.g. #294 - which means there does not need to be someone with write access to continually update the list.
We could also get some kind of automatic triaging bot to then assign these labels (project:error-index perhaps?). We could then have further human triaging - linking to relevant tests in GHC, tagging easy issues as low hanging fruit (e.g. "parse error: construct where it should not be").
I've done something similar at work (I proposed a new triage system, I got the go ahead, and I implemented it).
My concrete proposal: