Source-Locked Evaluation of Black-Box QUBO and Ising Heuristics: Public Evidence from Gset Max-Cut and BiqMac Certified-Value Checks

This record archives the public-safe evidence package for "Source-Locked Evaluation of Black-Box QUBO and Ising Heuristics: Public Evidence from Gset Max-Cut and BiqMac Certified-Value Checks." The paper presents a source-locked, append-only evaluation protocol for proprietary black-box QUBO and Ising solvers and applies it to B1Brain across locally controlled compute tiers. The public evidence snapshot includes completed solver rows across public-safe evidence lanes, a redacted Gset Max-Cut frontier extract, BiqMac certified-value parity checks, checksum manifests, public validation tables, and verifier/reproducer scripts. The archive is intentionally conservative. It supports claims about source-locked row accounting, public-safe evidence integrity, redacted per-run frontier objectives, checksum verification, and distributional reporting. It does not support claims of Max-Cut leadership, proven optimality for G72/G77/G81, timing superiority, throughput, energy efficiency, public rerunability of controlled baselines, or release of proprietary solver internals. Public GitHub evidence repository:https://github.com/bernie-b1stAI/b1brain-qubo-ising-evaluation GitHub release:https://github.com/bernie-b1stAI/b1brain-qubo-ising-evaluation/releases/tag/v2026.06.12