可微逻辑层误差累积的数学上界与正则化补偿机制

Run ID: run-20bb7cb0c3b6 Score: 0.795 Verdict: converged Generated: 2026-05-26