Spectral diagnostics and model integrity tooling for transformer systems

Geometric-brain-mcp

API, MCP, and Python interface for spectral diagnostics, comparisons, and audit workflows.

Open repo

Unitarity-lab

Research and audit engine for deep instrumentation, hidden-state analysis, and reproducible experiments.

Open repo

VAR

Enterprise beta monitoring layer for drift, anomaly, and integrity workflows.

Open repo

GUE Ensemble Target

r ≈ 0.603

The GUE (Gaussian Unitary Ensemble) r-statistic measures the average ratio of consecutive eigenvalue gaps in the model's hidden-state weight matrices. Random Matrix Theory predicts r ≈ 0.603 for systems with time-reversal symmetry breaking — the same distribution observed in chaotic quantum systems.

• r near 0.603: Eigenvalue spacing matches GUE prediction — normal spectral rigidity, no anomalous structure detected.
• r < 0.53 (Poisson limit): Eigenvalue repulsion breaks down — possible layer collapse, weight degeneracy, or uncorrelated blocks. Investigate with hidden-state analysis.
• r > 0.65: Stronger-than-expected eigenvalue repulsion — possible over-regularization, rank deficiency, or unusual training dynamics.

Berry–Keating k=1

k = 1

The Berry–Keating k-value characterizes coherence structure in the activation manifold geometry. k=1 corresponds to first-order coherence — the model's latent representations form a smooth, well-conditioned geometry consistent with stable information flow across layers.

• k = 1: First-order coherence intact. Latent geometry is well-conditioned. Model activations exhibit expected spectral correlations.
• k = 0: No detectable coherence structure. Activations are uncorrelated — possible dead layers, catastrophic interference, or meaningless residual stream.
• k > 1: Higher-order coherence detected. Can indicate strong geometric structure from fine-tuning, or — at high values — pathological entanglement between representations that should be independent.