Currently, if the server returns an error rather than results, then the TestNumber, Query and results are all missing from the .actual files. This leads to misleading logs from query_tester because if test 2 fails with an error, then the diffs will show

-2
-:sortFull {aggregate: "fuzzer_coll", pipeline: ... }
-[
- <expected results>
-]

and will indicate that the failing test number was 1 due to test 2 being completely absent from the .actual file.
I'll add some examples in the comments to this ticket.