Predictive Evaluation of an Externally Fixed Mass-Radius Normalization for Small Exoplanets

Author preprint, not peer reviewed. This Figshare item contains the complete paper and its Supplement, based on the frozen author manuscript v6.4. It is not an accepted manuscript or a publisher version of record. An externally fixed normalization can be evaluated predictively without estimating its central value from the target sample. We tested a mass-radius normalization fixed upstream at 6197.936 and 6155.396 kg m-3 in 145 selected 0.50-1.80 Earth-radius planets from 121 host systems. The primary estimand was the complete held-out-host joint log score, evaluated with grouped K10R5 and exact leave-one-system-out validation. After repair of a rare mode-initialization defect and independent same-node verification, all eight primary and strict-quality lower 90% paired-host-bootstrap bounds exceeded the declared -0.05 nat-per-host non-inferiority margin. Point contrasts ranged from 0.000570 to 0.016108 nat per host, but every interval crossed zero; compatibility with the margin was therefore supported without evidence of superiority. A continuous margin profile showed that the conclusion depends on allowing losses of about 0.005-0.022 nat per host across the eight cells. A fixed-normalization grid gave compatibility widths of 0.080-0.160 log10 density units at 0.05 nat per host, while the broader 0.50-nat specificity sets remained 0.347-0.694 units wide. Multiplicity-normalized host scoring preserved the same pattern: all lower bounds exceeded -0.05 and all intervals crossed zero. The coefficient is consequently compatible with the selected sample and scoring rule, but is neither superior nor uniquely identified. Registered radius-adequacy, specificity, cross-cosmology, and historical strict-quality limitations remain unresolved or failed. Supporting data-retrieval instructions, code, fixed configurations, public object/reference inventory and saved aggregate outputs remain in the separate OSF reproducibility record: https://doi.org/10.17605/OSF.IO/5BHM2 (release 4.0). The Figshare DOI identifies the paper; the OSF DOI identifies its supporting materials. This deposit does not replace or modify OSF. The reported positive result is bounded, post-confirmatory predictive compatibility/non-inferiority under the declared loss margin. It does not establish superiority, unique identification, a universal density, or a planetary/gravitational mechanism. The manuscript distinguishes the final posterior analysis from historical failed or inconclusive classifications. Funding: This research received no external funding. Competing interests: The author declares no competing interests. Licence after author-approved publication: Creative Commons Attribution 4.0 International (CC BY 4.0). The licence covers the author-created PDFs in this item; external works cited in them retain their own terms. See https://creativecommons.org/licenses/by/4.0/.