Estimate Stability — LessWrong