Isolated CPU Benchmarking of SAH-TriBVH for Narrowphase Collision Detection: Seeded Measurement Data (RUN2) and Manuscript Toolchain Scripts

This deposit contains the measurement data and manuscript toolchain for the paper "Isolated CPU Benchmarking of SAH-TriBVH for Narrowphase Collision Detection: Pruning Efficiency and Symmetric Query Analysis" (under revision, I.J. Engineering and Manufacturing, MECS Press). The primary artifact is the seeded RUN2 sweep: 230 rows covering three narrowphase engines (Brute Force, Moller-Trumbore symmetric, SAH-TriBVH) on the stock Stanford res4 meshes (Bunny 948, Dragon 11,102, Happy Buddha 15,536 triangles), object counts N = 2 to 20, ten samples per condition, deterministic per-sample reseeding (std::mt19937, base seed 1337 + sample index). OLD_vs_RUN2_report.md documents the full comparison between the originally reported sweep and the seeded re-run. Headline anchors: SAH-TriBVH reduces triangle pair checks by a factor of 1,984 against BF and 539 against MT (N = 2, Bunny); leaf-level pruning ratio 5.85% with 95% CI [5.79%, 5.92%] for N >= 5. The Python scripts reproduce all five manuscript figures from the CSV and record every revision edit as auditable document surgery. Mesh data courtesy of the Stanford 3D Scanning Repository.