Honest Compression Defense of the Spiral-Domain H-Pipeline: A 240-Study Pre-Registered Empirical Stress Test Against Ten State-of-the-Art Comparators with Foundation-Model, Cross-Tokenizer, Fine-Tuned LLM Downstream, Invertibility, Codebook-Portability, Fingerprinting-Robustness, and Energy-Proxy Coverage Including a Capable-Adversary Privacy Boundary Disclosure

This deposit accompanies the manuscript for Paper 16 of the spiral-domain encoder validation campaign series. It contains the pre-registration scope documents, runner code, summary verdicts, POC addenda, eight figures, and the arXiv-format manuscript for a 240-study pre-registered empirical defense of the H-pipeline compression claim against the strongest currently-available state-of-the-art comparators, executed in direct response to recent expert critique of SOTA-compression coverage and invertibility framing. The campaign comprises 240 pre-registered studies (1486–1725) across seven sub-campaigns (α, β, γ, δ, ε, ζ, η), all executed deterministically under PYTHONHASHSEED=0. The full campaign achieves 168 of 239 present per-study summaries SUPPORTED (70.3%), with all NOT_SUPPORTED outcomes reported verbatim and no retroactive threshold adjustment. Headline findings · Compression claim CONFIRMED THREE TIMES (Phase XXIX-δ §1583, ε §1614, ζ §1674) against ten named SOTA comparators at PhysioNet PTB-XL n=1500. The H-pipeline ranks first by BPE tokens per record: H-pipeline = 1 token, Chronos t5-small/t5-base = 1,537, ZFP (Lindstrom 2014) = 1,659, zlib = 2,333, gzip = 2,347, bzip2 ≈ 2,500, brotli ≈ 2,700, LZMA/xz ≈ 3,000, Gorilla (Pelkonen et al. 2015) = 4,924, naive raw-sample = 5,000. · Constant-output-size architectural property verified on real PhysioNet LTAFDB multi-hour Holter (Phase XXIX-ζ §1669–§1673, all SUPPORTED, variance = 0 across real 10s / 600s / 3600s ambulatory ECG; real-vs-tiled diff = 0). · Privacy claim honestly bounded (Phase XXIX-γ §1553 NOT_SUPPORTED, Outcome C). The §1510-style matched-privacy utility advantage observed at MIT-BIH n=48 does NOT generalize to PTB-XL n=1500 with capable LR/XGBoost/RF/MLP adversaries; Chronos is incidentally privacy-preserving at PTB-XL scale. · Fingerprinting robustness CONFIRMED: logistic regression and gradient-boosting MIA adversaries on 32-bin codebook frequency histograms both ≤ 0.55 AUC (Phase XXIX-η §1719/§1720). · Eight honest limitations pre-committed in §5: foundation-model coverage limited to Chronos (Gap F, second-attempt install failures honestly disclosed); fine-tuned LLM downstream AUC measured via Qwen frozen-embedding + LR proxy with H-pipeline at-or-better than Chronos head-to-head (Gap G PARTIALLY CLOSED); literal cross-tokenizer constancy fails at Qwen vs OpenAI BPE families, comparative advantage preserved across all five tested tokenizers (Gap H); GPU throughput unmeasured (L1); CPU-thermal-design-power Joule proxy explicitly NOT a wattmeter (L5); codebook portability quality shift > 3 dB (L3 PARTIALLY CLOSED); H-pipeline hard-lossy by design; multi-lingual extension out of scope. Related provisional patent applications All H-pipeline architecture validated in this manuscript is fully covered by the previously-filed Parents H/I/J/K/L. No new provisional patent application is required. · Parent H (U.S. Provisional 64/080,096, filed 2026-06-01; IoT-LLM pipeline; 62 claims) · Parent I (U.S. Provisional 64/084,807, filed 2026-06-11; federated multi-institutional; 50 claims) · Parent J (U.S. Provisional 64/084,817, filed 2026-06-11; codebook-centroid privacy hardening; 40 claims) · Parent K (U.S. Provisional 64/084,821, filed 2026-06-11; pre-encoder identity-axis projection; 50 claims) · Parent L (U.S. Provisional 64/084,827, filed 2026-06-11; real-time streaming H-pipeline + image/video modality; 70 claims) Companion deposits in the campaign series · Paper 8 — Phase XXI musculoskeletal-DT validation: 10.5281/zenodo.20466035 · Paper 9 — Phase XXII surgical-robot patient-pair DT: 10.5281/zenodo.20470763 · Paper 10 — Phase XXIII unidirectional release-only bioelectric DT: 10.5281/zenodo.20480183 · Paper 11 — Phase XXIV bidirectional closed-loop bioelectric DT: 10.5281/zenodo.20495976 · Paper 12 — Phase XXV IoT-LLM pipeline: 10.5281/zenodo.20499453 · Paper 13 — Phase XXVI federated multi-institutional: 10.5281/zenodo.20595834 · Paper 14 v3 — Phase XXVII real-time streaming: 10.5281/zenodo.20596597 · Paper 15 — Phase XXVIII image/video modality: 10.5281/zenodo.20596931 Reproducibility All 239 present per-study summary JSONs are bit-exact reproducible under PYTHONHASHSEED=0. The reproducibility corpus includes the H-pipeline-vs-comparator harness library (_common.py, _ptbxl_common.py), 13 deterministic runners (Phase XXIX-α through η), 6 per-sub-campaign FINAL aggregate JSONs, and 239 per-study summary JSONs. End-to-end campaign reproduction wall-clock: ~90–120 minutes on Apple M2 / 16 GB (CPU only) including PhysioNet record pulls. License Creative Commons Attribution 4.0 International (CC-BY 4.0). Commercial deployment of the H-pipeline architecture covered by Parents H/I/J/K/L requires a license to Fieldstone Analytics, LLC. Contact: randolphf@fieldstoneanalyticsllc.com. Citation Ferlic, R.J. and Ferlic, K.K. (2026). Honest Compression Defense of the Spiral-Domain H-Pipeline: A 240-Study Pre-Registered Empirical Stress Test Against Ten State-of-the-Art Comparators with Honest Limitations Disclosure. Zenodo. https://doi.org/10.5281/zenodo.20602959