An adaptive extension of HMC that automatically selects the trajectory length by detecting when the simulated path starts to double back on itself, eliminating the need to hand-tune the number of leapfrog steps. The default sampler in Stan and a core driver of the modern Bayesian workflow's practicality.