Teams often add automation metrics because a vendor template suggested them. Six months later the board is full and nobody can say why flaky jobs spiked last Tuesday.
A practical filter: write the decision sentence first (“We will change the deploy gate if recovery exceeds X”). Only then name the event and the properties that sentence requires.
Retired emitters deserve the same rigor. Mark them deprecated with an owner and a cut date so reliability reports do not silently break when platform cleans house.
Our audits usually find that half the volume comes from chatty debug agents and a retry loop left on. Cutting those frees attention for the dozen signals that actually steer delivery.