VALLY-Scan v9.5.1: Benchmark data for allosteric mapping on consumer CPUs

Benchmark data, trained BitPINN 1.58-bit model, and output results for the VALLY-Scan v9.5.1 framework. The dataset includes:- 62 per-residue 15D feature tensors (CSV) extracted from protein structures.- Trained ternary-quantized physics-informed neural network weights (PyTorch).- Feature scaler (joblib).- Benchmark outputs for three systems: SARS-CoV-2 RBD/ACE2 complex (6M0J, 791 residues, r = 0.68 in-sample, r = 0.73 leave-one-out), SARS-CoV-2 Spike trimer (6VSB, 2,905 residues, r = 0.80 in-sample, r = 0.72 leave-one-out), and alkane hydroxylase (5FWQ, 607 residues, r = 0.70 in-sample, r = 0.69 leave-one-out).- In-silico mutagenesis of TYR-508->ALA in 6VSB (Delta = 0.19, classified as mechanically silent). Full methodology is described in the associated manuscript. The core BitPINN and Lanczos engine source code is withheld pending patent review; a functional description sufficient for independent reimplementation is available in Supplementary Note 1 of the manuscript.