Verification and benchmark suite for faster-than-Nyquist geometry and detection

Le résumé fourni par la source

Verification and benchmark suite for "The Internal Geometry of Faster-than-Nyquist Signaling" (doi: 10.5281/zenodo.21724062). Every simulation, benchmark and finite check in that paper is reproduced by a script in this archive. Scripts are self-contained runners: `python .py` re-derives the quoted numbers and asserts them as gates, printing `[G# PASS]` per gate. A failed assertion means a printed claim did not reproduce. The JSON files are the recorded outputs of the runs behind the paper, so the archived numbers can be checked without re-running the campaigns. README.md maps every script to what it backs. The paper's evidence chain covers: the computer-assisted step of the Mazo-type threshold theorem, the geometry and threshold results, the cluster-aware sphere decoder and the cascade, the shared tree, the sublevel census with its walls and rim, and the signaling study behind the junction law. The archive also carries the shared infrastructure those campaigns run on, including the exact maximum-likelihood reference each graded cell is measured against, and a group of additional scripts the paper does not cite, retained as working material for continuing research. Requirements: Python 3.14 and NumPy 2.5 for all scripts; additionally Numba (campaign kernels), Matplotlib (figure modes), CVXPY with a conic solver (semidefinite row only), and PyTorch (auxiliary only).

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.