Cross-platform transfer of blood gene-expression models in Parkinson's disease — complete English reference edition of preregistration v3.1 and amendment v3.1.1

Complete English reference edition: https://doi.org/10.5281/zenodo.23406064 . This original preregistration record retains its original files and publication timestamp; the full English protocol, numerical documents, audit and translation table are available in that edition. The translation was published after the disclosed first GEO pilot, with no methodological changes. Preregistration protocol v3.1 for a cross-platform study of blood gene-expression models in Parkinson’s disease. Development uses PROBE/GSE57475 (142 records) and complete-case GENEPARK-TRO/GSE99039 (239 records); the external metadata-aware holdout is PPMI Project133 IR3-B38. A user-run local aggregate identifier audit dated 15 October 2026 reports 4,871 unique count-to-release-row links: 4,752 header/filename agreements and 119 exact header-only links. It reports 311 PD and 161 healthy-control baseline candidate records connected to portal metadata using sample identifiers, PATNO, visit, sex and historical labels. These are reported computational identifier links, not validation of biological donor identity or a finalized external analysis population; prescribed identity, QC and eligibility checks remain. The sole primary comparison is external AUROC(M5) minus AUROC(M4), using the same unpenalized demographic and six cell-marker baseline, with an additional penalized gene block only in M5. Training-only source-by-class weights, within-source macro-AUROC tuning, a paired stratified donor bootstrap, fixed transfer gates and an explicit baseline-only branch are prespecified. The package documents synthetic numerical validation, fit-level KKT checks, a justified solver amendment, and limitations of the two-source null simulation. These are implementation checks, not PD accuracy results. Metadata and linkage were audited before registration; no study model or PD/sex/age/cell-marker score has been evaluated in PPMI. The public files contain methods, code, public reference definitions and synthetic aggregate results, not individual PPMI data. Inconclusive outcomes are accepted in advance. The complete English reference edition is linked above. Metadata correction on 16 October 2026: author name and ORCID standardized; the identifier-linkage description clarified. The original v3.1 protocol files remain unchanged.