Your agent abstains on 8% of cases and every one becomes a human's problem. Design the escalation system.

The senior signal is treating escalation as a first-class product surface rather than an error path, and noticing that the escalation queue has its own capacity which the abstain rate is silently sizing. The reserve probe is the partial action: the agent did three things and then abstained, so the human inherits a case in an unknown state, and a design that hands over the question without the work already done is worse than no automation. Strong answers close the loop by feeding resolutions back as evaluation cases. Watch for anyone who routes by round robin, or who sets an SLA without checking the queue can meet it at peak.