Per-observation relative efficiency from the fit's chain structure
Source:R/worker.R
relative_eff_from_chains.RdMirrors brms' internal r_eff_log_lik() by deriving the chain id per draw
from posterior::as_draws_df(fit)$.chain and passing it to
loo::relative_eff(exp(ll), chain_id = ...). ll here is S x N (draws x
observations, matching brms' log_lik orientation): relative_eff wants the
draws dimension as rows so chain_id (length S) matches nrow(ll). Returns
one r_eff per observation (length N). Returns NULL when chain information is
unavailable (e.g. MockFitter or a fitter whose fit lacks a .chain
variable), in which case the caller falls back to r_eff = NULL.