The current iteration of this test tries to hand pick a manual winner first and then dances around the possibilities that stem from that. A better approach is to test all of the contenders so we can be sure there are no anomalies, and then select a manual winner that we know isn't already a winner.
Depends on D17695