Case Study: Testing Confirmation Bias — LessWrong