Bayesian Scenario: Snipers & Soldiers — LessWrong