As we ran the twist playoffs, we realized that the results were far too sensitive to the pod selection (and, apparently, the random number seeds). So, instead of running the playoffs as advertised, we randomly picked sets of six entries and put them together in pods. We then tracked the average place (one thru six) of each entry across these pods and when the averages settled down, we picked the three entries that had the smallest average place and declared those to be the winners.
Below are the results of each of the pods we ran (there are a total of 2,100 matches represented). We built enough image transcripts so that each participant in the playoffs would appear in at least two different pods with transcripts, and each participant would appear in a pod with each of the three winners at least once.