Papers, Benchmarks & Ethics

All OmniSciences publications, validation results, and our commitments to open science and responsible innovation.

Papers

1

The Lorentzian DeWitt Metric and the Pati-Salam Gauge Group

The DeWitt supermetric on GL+(4)/SO(3,1) has signature (6,4). The structure group SO(6,4) has maximal compact subgroup SU(4)×SU(2)L×SU(2)R — the Pati-Salam gauge group. Two-step RG running gives sin²θW(MZ) = 0.2312, matching observation exactly. The negative-norm sector is identified with the Higgs bidoublet.

DOI: 10.5281/zenodo.18860687 Accepted — Physics Letters B (2026) Interactive Figures
2

Fermion Representations from the Metric Bundle

The Clifford algebra Cl(R6)⊗C of the positive-norm sector produces spinor C8 = S+⊕S = 4⊕4̅ of Spin(6)≅SU(4). Combined with the negative-norm SU(2)L×SU(2)R, this yields exactly one Pati-Salam 16-plet per generation: (4,2,1)⊕(4̅,1,2). Hypercharge assignments follow from SU(3)×U(1)B−L branching.

3

Gauge Fields from the Gauss Equation of the Metric Bundle

The Gauss equation for the embedding X4 → Y14 yields the total curvature decomposition: RY = RX + gauge field strength + torsion. Einstein-Hilbert, Yang-Mills, and torsion terms emerge from pure metric bundle geometry with correct relative signs for the effective action.

4

Anomaly Cancellation in the Metric Bundle Pati-Salam Theory

All 13 triangle anomalies cancel in the Pati-Salam theory derived from the metric bundle. Mixed anomalies (SU(4)3, SU(2)3, gravitational, and U(1)B−L combinations) are verified computationally. Quantum consistency follows from the geometric Clifford algebra spectrum without requiring additional fermion families.

5

The Three-Generation Problem in the Metric Bundle Framework

NG = 3 from the Spinc index theorem on K3 with U(1)B−L line bundle. The base manifold is uniquely K3 (Hitchin-Thorpe + Ricci-flat + Rokhlin). The index computes NG = (c12 − σ)/8 = (8+16)/8 = 3, with no free parameters.

6

Quantum Mechanics from Finite Observation

The Born rule is the unique probability assignment preserving the Fisher-Rao metric on belief states. Fisher-Rao = Fubini-Study on quantum state space. Interference and complementarity emerge from blanket-mediated observations. Complex amplitudes are shown to be necessary and sufficient for bounded information processing.

7

The Complete Pati-Salam Effective Lagrangian from the Metric Bundle

The full effective Lagrangian with gauge couplings g4=gL=gR, Higgs bidoublet (1,2,2), and three fermion generations from K3. Key predictions: αPS = 27/(128π²) ≈ 0.0214 (7% from observed); sin²θW(MZ) = 0.2312 (exact match); Cabibbo angle sin(θC) = 1/√20 (0.75% error).

8

Localisation from the Free Energy Principle on the Metric Bundle

The Free Energy Principle applied to the metric bundle determines the gauge coupling αPS = 27/(128π²) from a variational principle. The soldering mechanism fixes Λbare with correct de Sitter sign. Localisation width from fibre Fisher-Rao geometry implies observable scale Lobs ~ 80 μm.

J

Independent Numerical Validation of Vibronic Nonadiabatic PCET Rate Theory: Head-to-Head Comparison with pyPCET

Independent implementation of multi-channel vibronic PCET rate theory validated against pyPCET with 0.1% agreement on BIP benchmarks. An initial 6.6× gating discrepancy is completely resolved (Vel = 1.75 vs 1.7 cm−1, 1.03×) through identification of three implementation-level errors. Monte Carlo UQ reveals 50–100% rate uncertainties.

Preprint: 10.5281/zenodo.19074557 Under Review — ACS Omega (ao-2026-03904b)
J

Riemannian Optimal Fingerprinting for Climate Attribution

Riemannian geometry on SPD covariance manifolds applied to climate forcing detection. All three major forcings (anthropogenic CO2, volcanic aerosol, solar) detected at p<0.01 on real observational data (HadCRUT5, GloSSAC).

Under Review — Journal of Climate (JCLI-S-26-00229)
A

Consciousness as Intrinsic Nature: Structural Idealism and the Formal Foundations of Mind

Philosophical paper proposing that consciousness is the intrinsic nature of physical structure. Develops Structural Idealism as a position in philosophy of mind, connecting Hoffman's conscious agent framework with Friston's active inference via the formal functor Θ: ConMB. Argues that the hard problem dissolves under structural idealism.

B

Conscious Agents and Markov Blankets: A Formal Correspondence

Technical paper establishing the formal correspondence Θ: ConMB between Hoffman's conscious agent theory and Friston's Markov blanket formalism. Proves compositional preservation (Θ respects ⊗), conditional independence from P·D·A factorization, inverse mapping characterization, and connections to Tononi's Φ (integrated information).

DOI: 10.5281/zenodo.18521824 Submitted — Journal of Consciousness Studies
J

Multi-λ Riemannian Features for Polarimetric SAR Land Cover Classification

Multi-scale geometric features from the DeWitt metric on SPD(3) applied to AIRSAR San Francisco quad-pol data. 96.73% overall accuracy with a single-parameter family of metrics. Few-shot regime: +9.3pp over Euclidean baseline at 5-shot.

Under Review — IEEE GRSL
WP

TR-2026-001: Riemannian Deep Learning for PolSAR Land Cover Classification

96.73% overall accuracy on AIRSAR San Francisco benchmark. +9.3pp improvement in few-shot regime over Euclidean baseline.

WP

TR-2026-002: Multi-λ Riemannian Regularization for Diffusion Tensor Imaging

76.4% classification accuracy vs 69.2% raw (+7.2pp) on Stanford HARDI benchmark. Patent #64/011,831.

WP

TR-2026-003: Geometric BCI via DeWitt Covariance Methods

+0.78pp improvement over pyRiemann baseline on 109 PhysioNet subjects (p=0.0048, Wilcoxon signed-rank).

WP

TR-2026-004: Riemannian Covariance Estimation for Portfolio Optimization

0% SPD failures vs 63% Euclidean. Conditioning validated at p<0.0001 (Fisher exact test, 5 synthetic universes). Patent #64/007,419.

WP

Geometry-Aware Materials Classification on SPD(6)

96.9% crystal system classification (vs 72.4% Euclidean) on 1,181 DFT-computed materials. 13,513× representation invariance improvement. 2,018 non-physical tensors flagged in Materials Project. Patent #64/004,047.

Benchmarks & Results

Reproducible results across all domains. All code and data publicly available.

Zero-Parameter Physics Predictions

SPD Manifold Operations — Latency

PCET Engine — 15 Enzyme Systems

BCI Classification — Motor Imagery

BCI — State of the Art Comparison

Domain-Specific Accuracy

Test Suite

Open Questions & Known Limitations

We believe in honest science. Here's what we don't yet know.

Physics

  • SU(2)R breaking mechanism undetermined — neither Coleman-Weinberg nor instantons produce the required hierarchy
  • Neutrino mass sector requires free parameter (Dirac Yukawa from b/a=0 gives mν too large)
  • CKM mixing beyond Cabibbo angle not derived (Wolfenstein A, ρ, η are fitted)
  • Flavour sector (b/a, Froggatt-Nielsen charges) needs dynamics beyond symmetric-space tree level

Applied Domains

  • BCI accuracy (+0.9pp) is modest vs deep learning methods (~96%); curvature features complement but don't replace SOTA
  • DTI curvature anisotropy not yet validated on clinical cohorts (MS, TBI, stroke)
  • PCET benchmark limited to 15 enzymes; broader validation needed
  • Portfolio optimization backtested on synthetic data; real market validation pending

Ethics & Licensing

OmniSciences is committed to open science, responsible innovation, and equitable access.

Open Research License

Perpetual, worldwide, royalty-free, irrevocable license for qualified users: academia, nonprofits, and government agencies. Non-commercial research and education always free.

Non-Aggression Covenant

No patent suits against universities, individual researchers, or nonprofits. Ever. If breached, patents automatically transfer to a public interest entity.

Humanitarian Pricing

For neglected tropical diseases, food security, and environmental remediation: pricing capped at direct cost + 15%. Ethics Advisory Board can waive entirely.

Revenue Reinvestment

≥10% of patent licensing + API revenue directed to a Public Benefit Fund supporting open science and education.

Irrevocability Clause

Open Research License survives acquisition, bankruptcy, or any change in ownership. Commitments cannot be revoked by future management.

Reproducibility

All computational results backed by 166+ automated tests. Code, data, and methods publicly available. Every paper's figures can be reproduced from the repository.

How to Cite

BibTeX
@misc{omnisciences2026,
  title={OmniSciences: Geometric Intelligence for Science},
  author={Austermann, Sloan},
  year={2026},
  howpublished={\url{https://omnisciences.io}},
  note={Open-source toolkit and research platform}
}