Wired up metadata in query results. First pass had workers looking it up locally from their chunk files. 400-500ms for 500 lookups. Not usable -- that's longer than the actual search.
Moved the lookup to the artifact server since it already has all the source data. Still too slow. The artifact server wasn't built for random-access key lookups, it was built for streaming bulk operations. Asking it to do point queries is like asking a dump truck to parallel park.
Need a third approach. Getting sick of this particular problem.