A pre‑specified Bayesian joint latent‑factor illness‑death model with two ordered transitions (unimpaired/p‑tau217− → unimpaired/p‑tau217+ → MCI/dementia). Cell‑type latent aging trajectories are estimated from plasma proteomic panels with biology‑anchored sparse loadings; convergence on amyloid and tau axes is modeled explicitly; APOE haplotype, APOE‑excluded cell‑type PRS, and a shared pleiotropic factor enter the trajectory submodel. Individualized time‑to‑p‑tau217 positivity and time‑from‑positivity to symptom onset are read out as posterior predictive distributions. Non‑parametric random survival forests and a stratified AFT model per Xu et al. serve as pre‑specified benchmarks