Physical Limits of Information Preservation Across Stellar and Cosmological Timescales: Reference Architectures, Recoverability Models, and Cosmological Bounds

We formulate a multiscale research framework for the persistence and recovery of classical information from 10^5 years through stellar evolution and conditional cosmological limits. The framework distinguishes encoded persistence, conditional readability and actual observer-mediated recovery; specifies passive, distributed and regenerative reference architectures; couples materials, optical readout, space hazards and common-cause reliability; and defines a reproducible numerical demonstrator. A deliberately hypothetical Poisson example exposes the sensitivity of billion-year survival estimates to hazard assumptions and independence, but is not a prediction for any material or orbit. The manuscript provides a layered self-describing readout protocol, testable experiments, unit and sign conventions, and a hierarchy of epistemic status for far-future scenarios. The cosmological portion bounds what may be physically conceivable without asserting that a passive object can survive until heat death. Synthetic software end-to-end validation of the coding pipeline is reported [C]; original experimental calibration and physical end-to-end validation remain outstanding. The contribution offered here is the framework itself — the reference architectures, the computationally verified coding pipeline, and the fully specified experimental protocol that would calibrate it — as a methods foundation for deep-time archival research, not a claim of demonstrated deep-time survival.