AgentBoundary
Certification is scoped to the system that was actually tested.
A defensible AgentBoundary result depends on stable prompts, models, tools, memory behavior, policies, architecture, and retrieval paths.
What you receive
Concrete outputs for a specific decision.
- Scope boundary explanation
- Change triggers
- Recertification decision points
- Customer-facing limitation language
- Recommended change-control practices
01
Changes that can invalidate a result
- Model swaps or material model configuration changes
- Prompt, system instruction, policy, or guardrail changes
- Memory behavior changes
- Tool additions, removals, or permission changes
- Architecture changes that alter authority, isolation, retrieval, or approval flows
- Major retrieval changes including new document sources, metadata policies, or tenant boundaries
02
Defensible language
A result should state what was tested, when it was tested, what scenario set was used, what versions were in scope, and what changes require reassessment.
Limits
Scope and assumptions stay explicit.
- This page explains limitation handling; it does not verify a specific vendor result by itself.
Next step