RGC v0.2: Verified 49-Core Logistic Substrate on Arty A7-100T — 3rd Extended Characterization (4M frames, ~159 h) with N=2 Cross-Run Determinism

This dataset contains the complete raw data and supporting analysis from a continuous 158.73-hour run of a 49-core Resonance Graph Core (RGC v0.2) implemented on a Digilent Arty A7-100T FPGA — the longest run in the project to date. The design executes the logistic map across a parameter sweep (r = 3.500 → 4.000) with Q1.23 fixed-point arithmetic. Key Results: Longest-baseline characterization and bit-exact cross-run determinism - 3,999,951 frames captured over 158.73 hours (~6.6 days) of uninterrupted, bit-clean operation; 81,632 lattice rotations.- Perfect iteration stability: Δn = 49 exactly across all 3,999,951 inter-readout intervals (zero anomalies), verified frame-by-frame over 49 cores × 159 hours.- Cross-run determinism (N = 2): this 4M run (cold-start of 2026-05-23) was compared bit-for-bit against an independent cold-start (2026-05-19, 1,209,644 frames) on the shared prefix. Comparison on the raw Q1.23 integer state, aligned on global iteration n and core_id: 1,209,644 / 1,209,644 frames matched, zero mismatches, 100.000000% agreement across all 49 cores. Confirmed independently by two separate comparison implementations.- Faithful reproduction of canonical logistic map dynamics, including period-4, period-5, period-3 (Sharkovskii window at Core 32), period-10, and broadband chaotic regimes.- Per-core invariant density, numerical Lyapunov exponents (approaching ln 2 at r = 4.0), symbolic entropy, RQA, and return maps confirm regime-specific behavior consistent with theoretical expectations.- Invariant measure converged: periodic cores lock by N ≈ 100k; slowest chaotic-transitional cores settled by N = 4M. Void convergence (first-half vs second-half Jaccard) mean 0.9704.- Core independence confirmed two ways: cross-core Pearson correlation at noise floor (off the low-r periodic block), and transfer entropy at the surrogate floor for all chaotic cores.- Clean 230400-baud UART telemetry; cumulative host-clock drift ~1.1 s over the full run (~7 ppm host-vs-FPGA offset, no bearing on the computation). Purpose: This work establishes the longest-baseline long-term-stable hardware logistic-map substrate to date and closes the determinism claim of the series: the substrate is bit-stable within a run and bit-reproducible across independent cold-starts (N = 2). No claim is made that the present data supports any specific theoretical prediction. The dataset is released as a verified engineering milestone and as the foundation for upcoming pre-registered experiments. Interpretation note (geometric visualizations): figures that wrap per-core state onto an S¹ × ℝ cylinder use the normalized state as an angular plotting coordinate; this is a visualization choice, not a physical dimension of the device. The associated signed-twist / cumulative-winding quantity is reported as an upper bound on angular drift in the return-map plane (a bound, not a detection), and the surface curvature is flat to the ~0.25% sampling floor (a null). These descriptors carry no physical claim about the hardware. Contents- Bifurcation diagram (4M, both runs overlaid)- Δn integrity heatmap and per-frame health (zero anomalies)- Cross-run determinism summary and figure (N = 2, bit-exact)- Cross-core correlation (raw and periodic-block-masked)- Cycle length per core (Q1.23 finite-state)- Invariant density grid and convergence-vs-N- Lyapunov exponent vs r- Per-core summary- Return maps and trajectories- Symbolic entropy vs r- RQA metrics and summary- Transfer entropy matrix and metrics- Persistent homology diagrams and H0 persistence- Signed-twist metrics (reported as bound) and cylinder-manifold visualizations- Timing analysis Context: This dataset belongs to the Tau Universe Framework (TUF2026) effort to explore hardware embodiments of resonance-graph computing. Future work will include pre-registered τ-imprint tests and extension to multi-board distributed resonance graphs. License: CC BY 4.0