Find the Metric That Will Be Gamed
Added 2026-11-03
Someone will optimise for any metric; decide the cost up front.
We are about to measure performance with a set of metrics (below). Do: 1) for each, how it is most likely to be gamed; 2) what gets sacrificed when it is gamed — quality, long term, other teams; 3) pair each with a counterbalancing metric; 4) identify the most dangerous one and why; 5) list signals that the metric improved while the work got worse. Do not just say the metrics are badly designed.
Site original
Chinese original
How to use
指标一旦成为目标，就不再是好的测量。
Try tweaking it
- Ask what evidence would overturn the judgement.
- Have it argue both sides with three reasons each.
What you should get
- A decision table: assumptions, evidence, counter-evidence
- A next action with an observable signal
Common mistakes
- Asking only "should I?" and getting equally valid answers both ways.
- Never asking for counter-examples, so you only hear support.
How to check the output
- It states what evidence would overturn the conclusion.
- Both sides are argued, with assumptions marked as such.
Where it works
- Good for: splitting a decision into assumptions, evidence and counter-cases, leaving the call to you.
- Watch out: ask what evidence would overturn the judgement, or you will only hear supporting reasons.