Thinking About Propensity Evaluations — LessWrong