Universal Intelligence Architecture (UIA) Dataset: Manifold Signatures and Behavioral Geometry Stress Tests

This dataset contains the complete experimental materials and raw telemetry used in the study:
The Universal Intelligence Architecture: Behavioral Geometry and Manifold Signatures in Large Language Models.

The repository enables full reproducibility of the reported results and includes:

• Raw multi-run stress test outputs (n = 3 independent runs per model)

• Baseline vs. UIA-conditioned measurements

• Per-metric deltas and variability calculations

• Aggregated summary tables

• AX–BY–CZ mechanical determination procedure

• Variability and unpredictability computation protocol

• Prompt suites (A1–A9 stress patterns)

• Documentation (README) describing structure and processing steps

Metrics include entropy measures (mean and P95), F1–F4 phase segmentation values, token counts, latency statistics, coefficient of variation (CoV), hedge rates, and self-reference markers.
Process metrics and persona metrics are classified according to the mechanical criteria defined in the included AX–BY–CZ Determination Procedure.
This dataset is released to support independent verification, replication, and extension of the behavioral geometry framework for model-agnostic AI system characterization.