The stored timeout applying across matching shapes is useful, but changing filter values can change the amount of work dramatically. A narrow reporting period and a year-long one may share a shape while needing very different execution budgets.

I would extend the timeout demonstration with several parameter values and record examined documents and completion rates, rather than calibrating from one representative request. That would expose whether a shared budget protects the service or rejects legitimate expensive variants. Keeping the stored setting and command-level value in the diagnostic output also makes the precedence rule easier to explain when a caller's longer timeout appears to have no effect.