How long must a single path of a Markov chain be observed so that the spectral gap can be accurately estimated from the observed data? I discuss some answers to this question from a few years ago, and time permitting, some new ideas.