Goodhart's Law
A useful metric can become less informative when strong pressure makes people or systems optimize the number instead of the outcome it was meant to represent.
Learn it → Apply it → Save as SkillWhen a proxy becomes a target, optimization can weaken the link between the proxy and the real goal.
Could this score improve while the outcome I care about stays flat or gets worse?
KPIs, incentives, education, AI evaluation, sales and performance management.
Fast overview
Charles Goodhart's 1975 observation concerned statistical relationships used for monetary control. The familiar wording, “When a measure becomes a target, it ceases to be a good measure,” is a later generalization associated with Marilyn Strathern's 1997 discussion of audit and accountability.
The practical lesson is not that targets always destroy measures. Optimization changes behavior, so the harder a proxy is pushed, the more carefully you should check whether it still represents the outcome.
Mental Model Compass
A number is tied to rewards, rankings, budgets or model optimization.
Gaming, threshold chasing, neglected quality and score-reality gaps.
How could someone raise this metric without improving the real outcome?
Concluding that measurement is bad or every target becomes useless.
Learn it progressively
State what success means before discussing its score.
Identify the indicator and why it currently represents the goal.
Ask what changes when pay, status, funding or automation depends on it.
Find ways the score can improve without equivalent improvement in reality.
Compare the metric with direct outcomes and revise when the relationship weakens.
Interactive Metric Stress Test
Examples
Heavy stakes on test scores can narrow teaching toward what raises the score rather than broader learning.
Judging sales only on booked revenue can encourage discounts or weak deals that hurt margin or retention.
Heavy optimization for a benchmark can improve benchmark performance without equal gains on wider user tasks.
Worked application
Goal: Resolve problems accurately and efficiently. Proxy: Average call duration. Pressure: Teams are rewarded for reducing it. Agents may rush difficult cases, transfer calls or discourage deeper troubleshooting. The speed metric improves while repeat contacts rise. Keep time as a diagnostic measure, but pair it with resolution quality, repeat-contact rates and direct customer outcomes.
Common mistakes
Metrics remain useful. Risk rises when a proxy carries strong control pressure and actors can adapt around it.
More metrics can add noise. Add complementary measures only when they protect an important dimension.
People can rationally adapt to rules without cheating.
A metric can lose information after incentives, technology or behavior change.
When not to use it
Do not invoke Goodhart's Law merely because a target exists. Some measures remain useful under targeting, especially when they closely represent the outcome, manipulation is difficult and reality checks remain visible. Use this model as a stress test for proxy-based control, not as a reason to abandon measurement.
Related models
Explains adaptation to rewards, penalties and constraints attached to a metric.
A metric represents reality. It is not reality itself.
Examines behavior induced by a target after the immediate effect.
Try It Yourself prompt builder
Save as Skill
Make Goodhart's Law reusable across capable AI systems.