Achievement is predictable, historical memory is almost not: a zero-leakage cross-era holdout on 5,350 eminent figures falsifies the innate-trait theory of greatness

A zero-leakage prospective holdout. Trained on 1,424 figures born before 1800 with only birth-fixed features (sex, civilisation, domain, innate g, neurotype posterior P_ND); predicted the 2,513 figures born 1800-1950 (n=5,350; 425 positive; bootstrap 95% CIs). For posthumous textbook eminence, T0 AUC=0.577 (.550-.606), +g=0.556, +P_ND=0.558; the independent net contribution of P_ND was Delta AUC=+0.0025 (random-label control 0.485; halo room 0.009). For contemporaneous top achievement, adding g raised AUC from 0.558 to 0.818 while P_ND again added ~0 (Delta AUC=-0.0043). Rolling cutoffs (1600/1700/1800/1850) converged to 0.56-0.58 as samples grew; administrative/inherited-power domain g-only AUC=0.357 (below chance, separate logic). The 0.82-vs-0.57 gap is variance absent from birth-fixed models (period, recording technology, politics, network amplification, luck, symbolic need). The null falsifies a trait-to-eminence/destiny narrative and relocates explanation to intervene-able post-birth processes (harness); residual full-biography halo, right censoring, and domain pooling are stated. Short report; labels are most-plausible hypotheses, not diagnoses.