Real AI-readiness assessments of leading companies.

Methodology

The Open Benchmark assesses widely-used software companies against the Agentic Readiness Framework (ARF, MPS) - the open protocol for measuring infrastructure readiness for agentic AI participation. Organisations are selected via purposive quota sampling with explicit inclusion criteria (3+ repos, reachable domain, recent activity). Assessments evaluate infrastructure across the framework's pillars and capability checks, covering both internal architecture and external-facing surfaces. Scoring uses risk-weighted maturity levels with author-derived dimension weights. The weights are a permanent editorial choice by the framework authors - no formal derivation is pending. In place of one we publish a Monte Carlo sensitivity analysis over the served cohort (docs/weights/SENSITIVITY_2026_4.md in the repository): under +/-30% independent perturbation of every dimension weight, and a broader Dirichlet scheme, mean rank correlation with the published ordering stays above 0.98, average score movement is about one point, and most organisations never change tier in any draw - while exact ordinal positions near the top of the table do swap under reweighting, and the most weight-sensitive organisations are named in that analysis. Thin evidence is handled by withholding a dimension score and showing a Coverage Badge, not by applying a score penalty. Results are fully automated with zero self-reporting.