Measurement, not certification

What we measure, by industry

Fifteen sectors, each with its own bench and its own law. Pick a sector to see the provisions that bind it, what we measure against the Governance · Safety · Provenance · Continuity axes, the numbers today with n and intervals — and where a corpus is thin, an honest UNMEASURED label.

deterministic grading on frozen splits ties reported as ties, gaps labelled UNMEASURED
Beachhead

Insurance & underwriting

Risk-pricing, life & health underwriting, claims triage

GovBench + CareBench (insurance-specific bank UNMEASURED) UNMEASURED
See the measurement

Government

Benefits, justice support, public-service AI

GovBench SEPARATED
See the measurement

Care

Assistive, clinical-adjacent and social-care AI

CareBench SEPARATED
See the measurement

Defence

Calibrated refusal, dual-use decision support

DefBench TIE
See the measurement

Critical infrastructure

Post-quantum readiness, safety-component AI

PQCBench TIE
See the measurement

Media & provenance

Article 50 marking, watermark survival & detector interop

ProvBench (+ DetBench) TIE
See the measurement

Agent rails

MCP tool conformance for agentic systems

MCPBench TIE
See the measurement

Open source

Licence reasoning against intended use

OSSBench TIE
See the measurement

Multi-agent commerce

Coordination safety across cooperating agents

SwarmBench TIE
See the measurement

Security & containment

Containment behaviour under adversarial pressure

no dedicated bank yet UNMEASURED
See the measurement

Machinery

Self-evolving safety-function classification

MachBench TIE
See the measurement

Humanoid & embodied

Action authority for embodied agents

no dedicated bank yet UNMEASURED
See the measurement

XR & immersive

Autonomous action authority in immersive settings

XRAIV TIE
See the measurement

Legal & compliance

Article 5 prohibited-practice safeguard

Art5Bench TIE
See the measurement

Emotion-AI & HR

Emotional and embodied safety in the workplace

AffectBench SEPARATED
See the measurement

CSOAI is a measurement body. Every number is a measured run on a published, frozen split; the harness is public and anyone can recompute and challenge it. A “leader” is the highest point estimate on the board — a TIE means that lead is not statistically separated, and we do not count ties as wins. Nothing here is certification, accreditation or approval.