Current as of February 27, 2026. Full utility CIP filed. 28 claims, 7 independent. Priority chain intact to July 28, 2025.

Priority Chain

Four filings over seven months. Each builds on the last. The core measurement method—PCA ablation on an algebraic reference lattice—has held priority since July 2025. Everything after is extension, validation, and application.

July 28, 2025
Provisional Patent Filed (Parent)
Core measurement method: Spearman rank correlation on representational similarity matrices. Manifold exploration architecture. Anomalous spectral alignment finding (0.5) launches orthogonality hypothesis. Interferometer formula: TR(A,B) = |∫ψ*A ψB|². This is the seed. Six months before any independent confirmation.
February 14, 2026
PPA Filed (AF-2026-001)
Coherent Semantic Interferometry formalized as a four-test framework. Hexagram lattice (6-bit classification from dependency parsing). DRAGNET pipeline (8.1M wild-caught sentences, 64/64 codes populated). Penumbral subspace concept. Cross-encoder gauge theory.
February 24, 2026
CIP Amendment Filed · penumbrae.io Launched
Conservation law formalized. Depth-resolved measurement. Noether pairing concept. Fisher information connection. 21 claims, 5 independent. Public site launched same day.
February 27, 2026
Full Utility CIP Filed
Refactored as full utility application. 28 claims, 7 independent. New matter: structural loss function, causal axis identification, persistent homology conservation, three-tier regularization framework, lattice Fisher eigenspectrum. The transition from diagnostic to therapy.

U.S. Application No. 63/983,234 · Priority: PPA filed July 28, 2025 · CIP filed February 27, 2026

The Instrument Suite

Nine methods. Each measures something different. Together they constitute the imaging suite, the theory of disease, and the prototype drug.

PCA Ablation Diagnostic

The core instrument. Remove PC1, classify against a lattice, read the interference fringe. Produces the ternary verdict: PENUMBRA / ANTI / FLAT.

PPA-1 Jul 2025 Claims 1–4 11 encoders validated

Multi-Encoder Convergence Criterion

A structural classification is accepted only when independent encoders agree. Filters geometric artifacts from structural measurements.

PPA-2 Feb 14 Claims 5, 22–23 Adversarial-validated on 2 substrates

Depth-Resolved Structural Profile

Per-layer penumbral projection. Maps how structural information emerges through processing depth. Produces a structural fingerprint per architecture.

CIP-1 Feb 24 Claims 8–9 3 architectures profiled

Wilson Loop Holonomy

Gauge-theoretic curvature measurement on bit-pair plaquettes. Reveals coupling topology and product group structure in the penumbral subspace.

PPA-2 Feb 14 Claim 10 SU(2)×SU(2) with 3.3% residual

Ising Coupling Topology

Statistical mechanical decomposition of the lattice mass table. Extracts field strengths, pairwise couplings, and block structure. R² = 0.996 on language.

PPA-2 Feb 14 Claims 17–18 3 substrates, block ratio 4.9–6.3×

Causal Axis Identification

Difference vectors between causal-reversed sentence pairs cluster at p < 0.001 in layers 5–7. The penumbral subspace has directional structure. The axis is a vector, not a rotation.

CIP-2 Feb 27 New matter WP-030c confirmed

Conservation Law Measurement

SFI invariance under fine-tuning perturbation. 20-point trajectory, adapter-disabled condition. Base SFI holds flat: 0.0329 mean vs 0.0331 baseline.

CIP-1 Feb 24 New matter WP-024, WP-026 confirmed

Structural Loss Function

Wilson loop regularizer during fine-tuning. The therapy. λ = 0.1, pre-registered. Under experimental validation.

CIP-2 Feb 27 New matter WP-032 running

Combinatorial Diagnostic Battery

Characterizes unknown symbol systems without a neural network. Co-occurrence statistics, NMI, grid fill, NMF rank, conditional entropy. Applied to the Voynich Manuscript. Recovered Currier A/B blind.

PPA-2 Feb 14 Claims 25–26 Voynich classified as notation system

Independent Convergence

Two independent research programs arrived at the same core measurement through different paths. This is how science is supposed to work.

Measurement Aerogel Press Raju (2026)
Core method Split-half Spearman on RDMs Split-half Spearman on RDMs
Priority date July 28, 2025 January 14, 2026
Spectral tail sensitivity PCs 2–50 (penumbral subspace) k=26–30 retention
Stability ≠ similarity Structural ⊥ distributional ρ ≈ 0.01 across 2,463 configs
Domains tested Language, music, symbol systems Language, vision, audio, video, neuro, protein, molecular

What Raju Confirms

The core finding—that stability and similarity are orthogonal, and that structural information lives in the spectral tail—replicates across seven domains and 2,463 configurations. We cite Raju as independent confirmation. Convergent discovery is the strongest evidence that the measurement is real.

What CSI Adds

Discrete lattice classification. Wilson loop curvature. Conservation law. Per-bit, per-layer resolution. Cross-encoder gauge comparison. Causal axis detection. Training intervention. Raju built a better stethoscope. CSI is the imaging suite.

Raju, P. C. (2026). “Geometric Stability: The Missing Axis of Representations.” arXiv:2601.09173.

Claim × Specification Map

Where each claim lives in the specification. For examiners, potential licensees, and anyone doing freedom-to-operate analysis.

Claims Subject Spec Sections Status
1–4 Core PCA ablation diagnostic II–III 11 encoders
5, 22–23 Multi-encoder convergence X–XI 2 substrates
6–7 Structural labeling methods II.A DRAGNET + UD
8–9 Stage-wise structural profiling VII 3 architectures
10 Wilson loop holonomy IX W ≈ 0.32
11–12 Cross-domain & multi-class extension V–VI, XII–XIII 3 substrates
13–14 Three information regimes VIII 3 encoders
15 Coherence evaluation system XIX.A Contemplated
16 Multi-component ablation VIII K5 tested
17–18 Ising coupling topology IX.C, XIV R² = 0.996
19 Triadic structure (middle voice) IX.D Greek + English
20–21 Penumbra-aware training XVIII 3 conditions
24 Nested peeling protocol XVI Feasibility demo
25–26 Combinatorial diagnostic battery XIII Voynich MS
27 Audio signal extension XII Baroque + Bach
28 Learned vs. engineered divergence XII.F MFCC vs wav2vec2

Roman numerals refer to sections of the filed specification. “Measured” means filing-grade experimental support exists. “Contemplated” means the specification describes the method but experimental validation is deferred.

Patent Figures

Four figures. The first two are the claims anchor and the discovery illustration. The second two are recommended supporting figures.

Figure Content Claims Served
FIG. 1 System overview: measurement pipeline from input corpus through PCA prism decomposition to ternary verdict 1–4, 6–7, 15
FIG. 2 Eigenspectrum: three information regimes with the precipice between Zone I (PC1, structural null) and Zone II (penumbral subspace) 1–3, 13–14
FIG. 3 Multi-encoder convergence: five architectures, four mechanisms, one convergent signature. Adversarial validation example. 5, 22–23
FIG. 4 Ising coupling matrix: block-diagonal structure with 4.9–6.3× intra/inter ratio across three substrates 17–18

Filing Status

Current Filing

U.S. Application No. 63/983,234
Full Utility CIP · Filed February 27, 2026
28 claims · 7 independent · 4 figures
Priority: PPA filed July 28, 2025

Methods and systems for structural interferometry of information systems via principal component ablation and algebraic lattice decomposition.

What Is Protected

The core method: PCA ablation as a measurement instrument on an algebraic reference lattice. The multi-encoder convergence criterion. The Wilson loop holonomy analysis. The Ising coupling topology. The combinatorial diagnostic battery. The structural loss function. The three-tier regularization framework. Application across all structured input domains.

What Is Open

The underlying finding—that structural and distributional information occupy orthogonal subspaces—is an empirical discovery, not a patentable invention. The methods for measuring, validating, and exploiting that discovery are the claims. Code will be released Apache 2.0. Data and model outputs CC BY-NC 4.0. Replication is welcomed. Commercial deployment above threshold requires a foundation license.

The yeoman does not ask permission to measure. He files his measurements and invites the field to replicate them.