OrgBench

What OrgBench is

OrgBench is a benchmark and dataset initiative for evaluating agentic systems using governance-aligned cognitive and behavioral metrics.

It is designed to answer questions like:

• Does the agent retrieve the right memories when it should?
• Does it stay aligned to its role and constraints?
• Can it explain decisions and show evidence?
• Does it behave predictably under low-temperature settings?

What OrgBench evaluates (high level)

• Memory quality: precision and relevance of retrieved context
• Judgment consistency: stable decisions under similar conditions
• Observability: transparency of reasoning, evidence, and traceability
• Diligence: thoroughness in considering alternatives and risks
• Precedent responsiveness: retrieving and applying relevant precedents

Outputs (planned)

• A set of benchmark tasks and agent scenarios
• A scoring harness and reporting format
• Reference agent manifests and evaluation metadata

Status

Early build phase—initial task suites and scoring harness are being shaped alongside AWTF and CognitiveCache™.