Privacy-Preserving Operations on Spiral-Domain Encoded Time-Series States: Anonymization, Aggregation, and Federated Learning with Composition Rules and Streaming Variants

This revised deposit contains the Paper 7 manuscript (PDF and DOCX) along with the supplementary reproducibility archive. The manuscript files were added in this revised version of the deposit; the supplementary ZIP file remains unchanged from the original deposit. Manuscript targets IEEE Transactions on Information Forensics and Security (under preparation). Coverage. 28 pre-registered studies spanning Phases XII-XX of the spiral-domain encoder validation campaign (privacy primitives foundation, edge-case stress, deployment realism, hardware context, composition + streaming, reviewer preemption, Tier-3 strengthening, plus surgical-RT latency). 84 hypotheses, 65 SUPPORTED (77%), 8 honest bounded negatives substantively interpreted. Substantive findings. Three architectural privacy primitives uniquely enabled by spiral encoder mathematical structure: Primitive A (per-subject angular phase-shift anonymization, k=10 cross-subject anonymity); Primitive B (cross-subject mean aggregation, 8.84 million-fold inversion resistance); Primitive C (federated AR(1) learning, exact 1-round convergence invariant to site count and heterogeneity). 5 deployment embodiments: multi-hospital clinical, multi-factory industrial, federated prosthesis fleet, surgical robotics RT privacy, cloud-scale parallel. Contents. Manuscript (PDF and DOCX of the paper itself); supplementary reproducibility archive containing: README.md (submission-package map and reproduction instructions); preregistrations/ (frozen pre-registration .md documents with literal-threshold decision rules); reports/ (per-study .md verdict reports against frozen rules + phase summaries); runners/ (deterministic Python runners under PYTHONHASHSEED=0); raw_data/ (per-study CSV outputs and JSON verdict blocks); figures/ (manuscript figures at 300 DPI + figure-build script); code/ (encoder source code). Reproducibility. Full validation pipeline is reproducible end-to-end under PYTHONHASHSEED=0 on a standard Python 3.9+ installation with NumPy 2.0+ and PyTorch 2.8+ (required for the learned-adversary autoencoder attack of Study 86). Reference machine: Apple Silicon arm64 (M-series), macOS 14. See README.md for per-study run commands. Methodological discipline. Every hypothesis was pre-registered with externally anchored decision rules frozen prior to runner execution. Zero post-hoc threshold adjustments were applied. Honest bounded negatives are interpreted substantively rather than discarded. Related companion archives. Paper 1 (10.5281/zenodo.20129137), Paper 2 (10.5281/zenodo.20138786), Paper 3 (10.5281/zenodo.20139171), and the corresponding Papers 4, 5, 6, 8 archives in this same Zenodo collection. Paper 8 (10.5281/zenodo.20466035) extends Paper 7 Composition III to musculoskeletal-kinematic clinical digital twin deployment with 21 additional studies and 91 hypotheses validated across CMU Motion Capture and KIMORE rehabilitation datasets including patient populations.