Cross-Domain Stress Testing of Unified Signal-Time-Authority Oversight: Independent C++ Toy Benchmarks for Tool-Use, Transactional, and Medical-Industrial Commitment

Abstract

This archive accompanies a synthetic toy diagnostic benchmark for unified signal-time-authority oversight. It includes a typo-checked manuscript PDF, an independent C++17 cross-domain simulator, raw and summary CSV outputs, Safety Slack calibration outputs, human approval latency stress results, figures, and reproducibility documentation. The benchmark covers simplified sandboxed tool-use, transactional, and medical-industrial action-bound domains. It tests whether runtime oversight performs best when monitoring cadence is sensitive to lower-tail intervention windows and action authority is preserved through cost-aware throttled gating. The results should be interpreted as simulation-based diagnostic evidence about qualitative framework behavior. This work is not real-world validation, not robotics or medical proof, not deployed tool-use safety evidence, and not a safety guarantee. License note: Manuscript, figures, and result files are released under CC BY 4.0. Source code is released under the MIT License.