Training Regime Day 17: Deflinching and Lines of Retreat — LessWrong