Very good feedback - thanks for showing me these screenshots. I’ve adjusted the algorithm and just deployed it. It should be a little stricter about matching stories up that shouldn’t match (i.e., false positives), while still maintaining the core clustering ability.