Controlled multi-region benchmarking

Conduct controlled multi-region benchmarking to rigorously evaluate Unbrowse and browser-based baselines across geographic regions using standardized cloud environments.

Background

The paper reports single-host, live-web benchmarks and acknowledges that absolute timings are geography-dependent. They indicate that a more controlled evaluation across regions is future work.

In the conclusion, controlled multi-region benchmarking is explicitly listed among the remaining open problems.

References

Formal economic analysis, incentive compatibility proofs, controlled multi-region benchmarking, and deployment-scale validation remain open problems; the present contribution is an architectural proposal with an implemented system and initial empirical evidence that shared route lookup can outperform redundant browser rediscovery on the evaluated tasks.

— Internal APIs Are All You Need: Shadow APIs, Shared Discovery, and the Case Against Browser-First Agent Architectures  (2604.00694 - Tham et al., 1 Apr 2026) in Conclusion (final paragraph)