Bayeswatch 3: A Study in Scarlet — LessWrong