A generator-matrix model quantifies the limited contribution of measured biomarkers to human mortality acceleration

Aging clocks and biomarkers are increasingly used as if they were the mechanism that drives mortality. But a mechanism, unlike a thermometer, must satisfy three conditions: it must account for the mortality signal, be causal, and be the layer that reverses when aging is reversed. Using public or provider-restricted de-identified data, we test all three. First (accounting), a Markov generator-matrix model with death as an absorbing state, fitted jointly to biomarkers and mortality in NHANES (n=23,512) and replicated in the Health and Retirement Study, assigns most of the Gompertz rise in mortality to a component latent to measured blood biomarkers (92.7%; 89.1% under a mean-field re-specification; 91.5% on replication). Second (causation), a positive-control-calibrated, two-platform cis-pQTL Mendelian-randomization design (UKB-PPP, deCODE) detects known causal proteins (LPA, IL6R) yet finds the measurable inflammatory, renal and growth-signalling markers null. Third (reversibility), at donor level reprogramming reverses a chronological clock (-9.7 and -22.4 yr), whereas a causality-enriched damage clock shows no detectable reversal while somatic identity is retained and moves only as pluripotency is approached. On these data the biomarkers meet none of the three conditions; aging measures appear to track mortality risk rather than its cause, a caution for their use as surrogate endpoints.