Why: The approved written design needs testable work items before changing model behavior or spending another sample.
Evidence: The plan maps all three design duties to existing code seams, RED-to-GREEN checks, a fixed combination gate and a separately authorized real-model exit.
Impact: Record stage-five review material without modifying product code, using paid models, or opening merge gates.
Refs: Agent-System-Documents#95; Agent-System-Dev#512