Counterfactual Calculation and Observational Knowledge — LessWrong