Uncertainty-Aware Ranking of Geroprotective Interventions from Public Lifespan Data: A Sentinel Cross-Species Benchmark

This is a secondary analysis of a sentinel public-record extract. It is not an exhaustive DrugAge meta-analysis, a clinical study, or a treatment recommendation. Public longevity databases contain heterogeneous observations collected across species, strains, sexes, doses, treatment windows and experimental laboratories. Ranking interventions directly from the largest reported lifespan extension therefore rewards extreme and weakly replicated records. We present a reproducible sentinel benchmark for evidence-aware prioritization of geroprotective interventions. The benchmark contains 58 explicitly enumerated records from an open DrugAge snapshot, representing 15 compounds, 24 independent compound–publication–species units and seven model organisms. Records were first aggregated within publication and species to reduce pseudo-replication from dose and sex strata. Candidate-level evidence was then represented by a capped effect component, a Wilson lower bound on the fraction of significant-positive study units, and study/species breadth. We combined these features in a transparent ranking and separately applied a non-compensatory gate requiring replication across at least two studies and two species, with positive study-level effects throughout. Aspirin, metformin, fisetin and curcumin occupied the top four positions in all 5,000 plausible score-weight draws, but the underlying evidence included unusually reported doses, small numbers of independent studies and large cross-context effects. Five compounds passed the strict gate, including 2,4-dinitrophenol, illustrating that lifespan evidence alone cannot substitute for safety, pharmacokinetic or mechanistic assessment. Rapamycin ranked below several cross-species candidates because the sentinel extract contained replicated mouse evidence but no second species, despite consistent positive mouse effects. The central result is therefore methodological: effect magnitude, reproducibility, cross-species breadth and safety must remain separate decision dimensions. This benchmark is not an exhaustive DrugAge meta-analysis and isnot a clinical recommendation. It provides a compact, auditable test bed for future transcriptomic, toxicological and virtual-tissue intervention models.