History-Dependent Response, Predictive Transfer, and Limits of Identifiability: Failure-Preserving Mathematical and Experimental Benchmarks from C19 to P92 Steel

A failure-preserving synthesis of a finite C19 boundary-history model, retrospective cantilever tests, and P92 steel creep-fatigue data from three independent specimens. We distinguish (i) forward history causality, (ii) external state/parameter identifiability, and (iii) added prospective predictive value over properly calibrated conventional models. A conventional two-branch generalized Maxwell baseline reached within-specimen RF NRMSE 0.02224; C19 noisy/unknown-gain readout and E9A/E9D/E9G joint transfer gates failed. Slow conventional modes improved separately calibrated long-time tails but a shared parameter test failed. No BIG-specific physical incremental prediction was established, and B42 SCIENCE NONE is retained. This record contains the English preprint, Japanese reference translation, and an author-generated protocol/code/audit supplement. Original BAM/Bristol source datasets are NOT redistributed; cite their originators. Unreviewed author preprint. No independently replicated physical experiment or full independent end-to-end software rerun. Manuscript revision 2026-11-05. English authoritative, Japanese reference translation.