Computes a lightweight initial-positive-sequence effective sample size, Monte Carlo standard error, and lag-one autocorrelation for saved scalar traces. PL fits include the total ranking log likelihood; block models also include their occupied-cluster traces.