# Valstorm Agent Runtime: Multi-Axis Scientific Benchmark & Ablation Study

> Reproducible 6-axis ablation study (UK AISI Inspect AI, Terminal-Bench 2.0, tau-bench pass^k, GAIA/SWE-bench) demonstrating a 64.8% token reduction (2.8x lower cost) via profile-scoped tool registries, 1.53x-3.00x parallel subagent swarm speedup via background=True, and 100% pass^k reliability with iterative ReAct loops and procedural skills.

- **GitHub Repository:** https://github.com/Valstorm-Corp/valstorm-agent
- **License:** PolyForm Noncommercial License 1.0.0 (https://polyformproject.org/licenses/noncommercial/1.0.0)
- **Full Study:** https://www.valstorm.com/research/agent-runtime-benchmark-harness