Sedimentary cores from estuarine and marine environments contain shelly records of past communities. However, using these to develop time-lines of ecological change for the last several millennia to centuries -- or even using death assemblages from grab samples to characterize "modern" diversity -- requires quantifying the reliability of species abundance data (e.g., differential destruction and post-mortem transportation of species) and the magnitude of time-averaging (memory of past generations). Observational and manipulative experiments, direct dating of death assemblages, meta-analysis, and modeling now permit quantification of many aspects of data quality and reveal underlying controls and tradeoffs. For example, "live-dead" comparisons of shelled mollusks, where counts of living individuals sieved from quantitative grabs are compared with counts of dead individuals in the same samples, indicate that average "dead richness" is ~25% greater than a single live census, reflecting the accrual of rare species during time-averaging (114 subtidal datasets; some shells are several thousand years old, but the vast majority are <a few hundred yr). Species relative abundance is quite robust: rank-order agreement of live and dead species lists is 0.31 (richness-weighted meta-analytic average Spearman rho of 173 comparisons), comparable to "live-live" agreement when two successive censuses of the same habitat are compared (rho = 0.34; N = 1732 comparisons). If only "pristine" habitats are considered, average live-dead rho increases to 0.38. Live-dead agreement declines significantly with increased anthropogenic eutrophication, bottom-trawling, or other human impacts, owing to dead composition lagging behind the shifting community baseline (rho = 0.24; taphonomic inertia). Poor live-dead agreement thus carries a strong signal of recent community change. Death assemblages represent time capsules that, because they time-average across short-term volatility, capture the pre-impact community composition as well as single live censuses. These are generally encouraging results for a variety of ecological analyses from molluscan death assemblages.