Adaptive Coherence — LessWrong