ORF Playbook
Build the evidence needed to operate, investigate, and govern generative and agentic AI systems responsibly.
The Observability Requirements Framework (ORF) defines a risk-tiered set of logs, traces, metrics, and review evidence. This playbook turns that framework into a practical implementation journey for product, engineering, risk, and operations teams.
Follow the playbook in four sections
The four sections below contain the six implementation stages: Understand, Scope, Design, Implement, Verify, and Operate.
-
Start here
Understand the outcome, audience, and approach.
-
Plan your implementation
Profile the system, determine risk and scope, select requirements, and design telemetry.
-
Implement ORF
Build the trace substrate first, then add investigation and analytical capability.
-
Operate and improve
Use telemetry in incident response, reviews, and continuous improvement.
What ORF gives you
- Reconstruct decisionsConnect prompts, outputs, retrieval, tools, and traces.
- Investigate incidentsRecover what happened, when it happened, and why.
- Review control operationUse measurable evidence and audit trails to assess whether controls were invoked and worked as intended.
- Improve continuouslyTrack performance, safety, reliability, and user signals.
Find what you need
Use this as a playbook, not a checklist alone. Begin with Start here, then use the reference section when you need the full framework definitions.