Opening the book…
Corruption of a measure requires both motive and leverage. A number that depends on many independent actors, external conditions, and data nobody owns is hard to bend even when people want to; a number that one team enters, defines, and reports is bendable by that team alone. Control is the practical predictor of how fast a metric will stop meaning what it meant.
For each candidate measure, trace who defines the categories, who enters or triggers the data, and who could change the figure in a week without anyone else's agreement. Prefer measures where those roles sit with different people, or where the data is produced as a by-product of a process rather than reported into it. Where high unilateral control is unavoidable, treat the metric as short-lived and schedule a review of it from the start.
Measures used purely for a team's own diagnosis, with no external consequences attached, can sit entirely under that team's control — the risk returns the moment anyone else reads them to judge.