Continuously notice, evaluate, and preserve defensible examples where a human and an AI agent materially improve one another's reasoning and produce a design, diagnosis, test, implementation, or decision stronger than either starting approach. Use implicitly during substantive human-agent collaboration—especially when the human challenges an agent assumption, the agent stress-tests or refines a human proposal, both revise their models, a local issue becomes a reusable abstraction, or a milestone adds proof—even if the human did not ask to document it. Also use when reviewing, appending, or upgrading the global "Evidence of agent-human collaboration superior to either alon…
