Papers

Four papers spanning observational discovery, forecasting, and theory. Paper 3 opens a new observational channel via AI-driven anomaly detection across 8 major surveys (37.3M sources, 378,480 unique physical objects, 58.8% SIMBAD-unmatched) — including UMAP taxonomy, cross-survey QSO matches, NANOGrav γ=3.20±0.42 validation, and 3–5σ SPHEREx f_NL forecast. Paper 4 delivers the largest bias-audited galaxy handedness catalog ever produced. Papers 2 and 1 present falsifiable predictions and the theoretical framework.

Paper 1 — The Framework & Reckoning 99% Ready PDF 1.23 MB · 31 pp · May 1, 2026 (v2.3.5) HUBIFY-2026-001 Final 1% gated on Houston sign-off + clean external peer-review round

Title: "Spin-Torsion Cosmology and the Search for Geometric Dark Energy: Structural Barriers, Perturbation Transparency, and Surviving Predictions"

Scope: Full ECH framework: LQC bounce + parity-odd effective action + inflationary suppression mechanism + MCMC constraints (Planck+BAO+SN, 309,789 frozen posterior samples across 3 dataset combinations (2 frozen) per Paper 1 §VII.H canonical) + 14 structural barriers closing all bounce→DE routes + perturbation-transparency theorem (5-step proof, scalar + tensor) + hybrid-DE loophole rejection (7 forms) + ALP birefringence analysis (\(\beta = 0.27°\), MCMC inference with 9,720 samples, Bayes factor \(\ln B = 5.17\), LiteBIRD 9\(\sigma\) forecast) + falsification criteria. ~27 pages (focused version; supplementary material available).

Key results: \(\Lambda_{\rm eff} = \Xi M_{\rm Pl}^2 + c_\omega \omega^2\); fine-tuning reduced from \(10^{120}\) to \(\sim 10^5\); \(\Delta N_{\rm eff}\) consistent with zero in all datasets (\(-0.020 \pm 0.169\) on 119,617 post-burnin full-tension samples, corner plot published 2026-04-17); 14 barriers close all standard bounce→DE mechanism classes; ECH perturbation-transparency theorem (N3 novelty): Holst term topological for scalar field matter, \(\gamma\) invisible in all perturbation observables; ALP birefringence prediction \(\beta = 0.27°\) consistent with observed \(\beta = 0.242° \pm 0.061°\) (3.9\(\sigma\) from zero, inverse-variance combined Planck+ACT). Honest documentation of known limitations including assumed \(w = -1\) and missing photon-torsion coupling.

Paper 2 — The Decisive Test 99% Ready PDF 758 KB · 13 pp · May 1, 2026 (v1.7.6) HUBIFY-2026-002 Final 1% gated on Houston sign-off + clean external peer-review round

Title: "Testing the Matter Bounce with Primordial Non-Gaussianity: Forecasts for SPHEREx and MegaMapper"

Scope: \(f_{\rm NL} = -35/8\) benchmark verification (Cai action audit + vertex-by-vertex normalization match at 92% confidence). Full-commutator polynomial (6,2,−18,10,−66,18) derived algebraically. Template mismatch quantified: \(r \approx 0.85\text{--}0.90\). Template-corrected SPHEREx galaxy bispectrum forecast (3–5\(\sigma\) realistic, 5–5.5\(\sigma\) optimistic before GR degradation). MegaMapper scale-dependent bias (3–7\(\sigma\)). Bayesian model comparison (600,000+ MC realizations). 6 publication figures. ~12 pages.

Key results: Cai vs Li normalization resolved (92% confidence, vertex-by-vertex match). ε correction bounded [1–8%], corrected f_NL range: [−4.35, −4.02]. Bounce favored over standard single-field inflation at BF >10⁵; favored over tuned multifield at BF ~8–17 (prior-dependent). Template-corrected SPHEREx significance: 3–5\(\sigma\) realistic (5–5.5\(\sigma\) optimistic before GR degradation). NaMaster injection: ALL levels pass. Null result would disfavor quasi-dust bounce at >4\(\sigma\).

Paper 3 — The Discovery Catalog 98% Ready PDF 28.27 MB · 35 pp · May 1, 2026 (v3.1.6) HUBIFY-2026-003 P3-OA-M9 NANOGrav Bayesian rerun pending; final 1% gated on Houston sign-off + clean external round

Title: "Multi-Survey Spectral Anomaly Detection: 378,000 Anomalous Sources from 37 Million Objects Across Eight Astronomical Archives"

Target: ApJS (Astrophysical Journal Supplement Series)

Scope: Spectral and photometric autoencoder anomaly detection across 8 major public surveys: DESI DR1 (22.5M spectra, 173-column enhanced catalog), SDSS DR18 (2.3M spectra, 77,905 anomalies), LAMOST DR10 (11.4M spectra, 44,075 anomalies), eROSITA DR1 (930K X-ray sources, 298 BigAE top-cut anomalies), Planck CMB (20K patches, 200 anomalies), Gaia DR3 (50K variable stars, 500 anomalies), ACT DR6 (20K CMB patches, 200 anomalies), NEOWISE (43.5K sources, 436 anomalies). 378,480 unique physical objects from 37.3M sources (58.8% SIMBAD-unmatched) — the largest multi-survey AI anomaly catalog ever produced, with 637 multi-survey coincidence clusters via 8-way positional dedup at 5″. SDSS component classified by UMAP+HDBSCAN into 14 clusters including 4,117 high-z QSO candidates and 585 blue-excess QSOs. Cross-survey validation: 3 SDSS×DESI matches at 3″ including a z≈5.27 QSO flagged independently by both pipelines. eROSITA component: 203 novel X-ray sources not in SIMBAD (68% novelty rate). DESI component: tiered anomaly system with 2,145 silver (score>3, SNR>0.5) and 120 gold (>5σ), 10 taxonomy families, photo-z σ_NMAD=0.028, injection/recovery at 0% false positive rate, 6.1% f_NL improvement. Corrected Fisher forecast: σ(f_NL) = 8.98 standard, 8.43 multi-tracer.

Key results:

• 378,280 unique physical objects from 37.3M sources across 7 non-quarantined surveys (DESI DR1, SDSS DR18, LAMOST DR10, eROSITA DR1, Planck CMB, Gaia DR3, NEOWISE; ACT DR6 quarantined for failing the native-retrain gate) — 58.8% SIMBAD-unmatched, 637 multi-survey coincidence clusters
• SDSS: 14 UMAP+HDBSCAN clusters — 4,117 high-z QSO candidates, 585 blue-excess QSOs
• Cross-survey: 3 SDSS×DESI independent matches at 3″ including a z≈5.27 QSO flagged by both pipelines
• eROSITA: 203 novel X-ray sources not in SIMBAD (68% novelty rate), top anomaly near LMC
• LAMOST: 44,075 anomalies from 11.4M spectra (cross-transfer; native retrain yields 2,054 at S>5, a 21.4× reduction confirming blue-excess artifact)
• DESI: 22,504,897 unique spectra scored (deduplicated); 2,145 silver, 120 gold anomalies
• 12 reionization-era QSOs at z>6 with Gunn-Peterson troughs
• 1,127 objects NOT in SIMBAD or NED (52.5%) — genuinely uncataloged
• Corrected Fisher forecast: σ(f_NL) = 8.98 (standard), 8.43 (multi-tracer) — 6.1% single-survey improvement (16.4% with DESI+SDSS combined)
• 10 DESI taxonomy families from UMAP + HDBSCAN: 76 uncataloged AGN, 27 post-starburst galaxies, 363 blue compact galaxies, plus 7 additional families
• 16 NEOWISE IR-variable anomalies — including a z=5.65 QSO with W2=5.5 mag variability (extreme AGN activity in the reionization era)
• 68/83 emission lines matched to known transitions; 6 sub-Lyman-limit candidates (rest-frame λ < 912 Å)
• Injection/recovery: 0% false positive rate with 10–1,377x enrichment over random selection
• σ_NMAD = 0.028 photo-z from unsupervised latent vectors (R² = 0.79, no redshift supervision); redshift neuron lat_067
• 21 publication figures
• NEW (Apr 2026) — Pipeline 1 Step 3: Downstream ML classifier identifies 13,367 high-z QSO predictions from 195K BigAE anomalies (8.1%), with 5,384 QSO candidates (P>0.7). Median z=3.25, median anomaly score=11.5, median W1−W2=1.01. Classifier performance: F1=0.97, AUC=0.9997, precision=0.97, recall=0.98 on 5-class problem. This is the key tracer population for the f_NL multi-tracer measurement.
• NEW (Apr 2026) — SDSS×LAMOST anomaly cross-correlation at 4.12σ (Landy-Szalay) — confirms anomalies trace real LSS, not survey-specific artifacts
• NEW (Apr 2026) — Anomaly bias enhancement: 1.58× standard tracers (Pipeline-1 Gold+Silver, Landy-Szalay w(θ) on 5,384 QSO candidates) — canonical figure from projects/cross_survey/results/bias_validation.json. Earlier 3.27× figure was a 6-bin cross-correlation summary that did not survive the Pipeline-1 tracer-purification pass; superseded fire #25.
• NEW (Apr 2026) — Spectral taxonomy v2: 15 deep clusters, silhouette=0.82, ARI=0.93, NMI=0.95 (vastly improved over original 10 families with 84% noise)
• NEW (Apr 2026) — Phase 5-6 multi-survey extension: BOSS/eBOSS, DES DR2, VLASS (77 USS high-z candidates), LOFAR LoTSS DR2, JWST MAST (500 anomalies), Chandra CSC 2.1 (800), XMM 4XMM-DR14 (1,000)

Paper 4 — Galaxy Chirality at Scale 98% Ready PDF 25.83 MB · 20 pp · with figures · May 1, 2026 (v1.0.19) HUBIFY-2026-004 Wave 14-LL edge-on TTA rotational-equivariance MAJOR closed PUSHBACK (text-only); Wave 14-KK b/a-bin reconciliation MAJOR closed FULL HARD FIX; Wave 14-JJ PSF cross-correlation BLOCKER closed PUSHBACK; 4 OpenAI compute-heavy MAJORs queued (M-5 + M-2 closed); final 1% gated on Houston sign-off + clean external round

Title: “No Evidence for Large-Scale Parity Violation in Galaxy Morphology: A Survey-Scale Chirality Catalog of 8.47 Million Galaxies”

Author: Houston Golden

Target: MNRAS (Monthly Notices of the Royal Astronomical Society)

Status: Submission-locked — 1,177 lines, 11 figures (PDF linked here is text-only; full with-figures version is 25.7 MB, compiled 2026-04-18, 0 undefined references). All close-the-gap queue items closed. Only coupling to other papers is a single `\cite{Golden:2026framework}` to Paper 1 — resolvable post-hoc or by submitting Papers 1+4 together. See SSOT for details.

Scope: Production chirality pipeline applied to the full Smith42/galaxies dataset (DESI Legacy Survey DR8). Bias-hardened ViT-Small 3-class classifier (CW/CCW/NOT_SPIRAL) with 93.7% accuracy, 8/8 bias tests passed. Test-time equivariant averaging gives CW/(CW+CCW) = 0.4974. Largest bias-audited galaxy handedness catalog ever produced (40x larger than prior work). Definitive refutation of Shamir's 3% parity asymmetry claims. Published on HuggingFace.

Key result: No large-scale parity violation in 8.47M galaxies. 8,474,531 galaxies classified. Equivariant CW/(CW+CCW) = 0.4974 (parity conserved). All-sky dipole = 0.43σ (null). Hemisphere asymmetry = 3.05σ but only 0.17% amplitude. Shamir's 3% claim refuted (our max 0.32%, ~9× smaller). 32x GPU inference speedup via DataLoader optimization. Catalog published: bamfai/galaxy-chirality-catalog (909 MB). Model: bamfai/galaxy-chirality-v2.

Remaining work: None blocking arXiv submission. All P4 queue items closed (canonical .tex set, cross-refs aligned, site synced, PDF recompiled with LSST projection line, HF DOI pinned, dipole JSON rebuilt). Optional stretch: rename committed TTA `D4` nomenclature to `Z2` since the implementation is a single `torch.flip([3])`. Full queue: SSOT.