Goodhart\'s Law
Goodhart's Law is the economic adage, formulated by the British economist Charles Goodhart, that states: When a measure becomes a target, it ceases to be a good measure. First articulated in the context of monetary policy — where attempts to control the money supply as an intermediate target led to its breakdown as a reliable indicator of economic activity — the law has since become a general principle of systems theory and institutional design, describing how the act of optimizing for a proxy destroys the proxy's informational value.
The law is not merely an observation about human ingenuity in gaming metrics. It is a structural theorem about the interaction between measurement and behavior in complex systems. Once a measure is announced as a target, agents in the system reorient their behavior to optimize the measure, often at the expense of the underlying goal the measure was intended to track. The measure and the goal decouple, and the system enters a regime of performative measurement — where the metric no longer describes reality but actively reshapes it in ways that conceal the very information it was designed to reveal.
The Logic of Gaming
Goodhart's Law operates through several mechanisms:
Substitution. When a proxy is used as a target, agents substitute behavior that improves the proxy for behavior that improves the underlying outcome. In education, teaching to the test improves test scores without improving understanding. In medicine, optimizing for patient-satisfaction scores may lead to overtreatment or avoidance of difficult conversations. In software, optimizing for lines of code written produces verbosity, not functionality.
Distortion. The measure itself becomes an object of manipulation. Citation counts can be inflated through coercive citation. Crime statistics can be reduced by reclassifying offenses. Customer service metrics can be gamed by disconnecting calls prematurely. The distortion is not always intentional or conscious — institutional incentives shape behavior more powerfully than explicit instructions.
Displacement. Resources and attention shift from unmeasured but important activities to measured but less important ones. A university that prioritizes research publications over teaching will see teaching quality decline, even if the teaching was never measured and therefore never formally degraded. The displacement is invisible because the neglected domain lacks a metric to register its loss.
Emergent complexity. In large systems with many agents, the aggregate effect of individual optimization is often unpredictable and harmful. Each agent responds rationally to the incentive; the collective result is irrational. This is the connection between Goodhart's Law and the tragedy of the commons, the principal-agent problem, and Campbell's Law — the sociological observation that the more important a quantitative social indicator is for decision-making, the more subject it is to corruption pressures.
Systems-Theoretic Reframing
Goodhart's Law can be understood as a failure of observability in control systems. In control theory, a system is observable if its internal state can be inferred from its outputs. Goodhart's Law describes what happens when the act of control — the feedback loop that uses the output to adjust the system — alters the dynamics that produced the output in the first place. The controller's model of the system becomes invalid because the controller's own actions have changed the system. This is not a failure of control design but a fundamental limit on the coupling of measurement and intervention.
The cybernetic reframing is sharper: a system that uses its own output as input enters a reflexive loop. In stable regimes, reflexivity is benign. In unstable regimes — where the gain of the feedback loop exceeds the damping capacity of the system — the loop produces oscillation, amplification, or collapse. Goodhart's Law is the social-scientific name for a feedback instability that engineers have studied for decades under different names.
Countermeasures and Their Limits
The standard response to Goodhart's Law is to improve the measure — to design metrics that are harder to game. But this is often futile. The problem is not the quality of the measure but its function as a target. Any measure, however well-designed, will be gamed if the stakes are high enough and the monitoring capacity is limited. A better response is to decouple measurement from reward: to use metrics for diagnosis rather than evaluation, for exploration rather than accountability. But this requires institutions capable of tolerating ambiguity — a capacity that is itself scarce under competitive pressure.
Another response is to use multiple, diverse measures rather than a single target. This is the logic of dashboard governance and balanced scorecards. But diversity of measures does not eliminate gaming; it merely displaces it. Agents optimize the weighted combination, and if the weights are known, the combination becomes a single target with a more complex formula. The Lucas critique in macroeconomics makes a parallel point: policy rules that are effective when unanticipated become ineffective once anticipated and exploited.
Goodhart's Law is not a problem to be solved by better metrics. It is a theorem about the limits of centralized optimization in systems composed of intelligent, adaptive agents. The deeper lesson is that any system that rewards what it can measure will eventually measure what it can reward, and the two will drift apart until the system operates in a fantasy of its own construction. The only sustainable countermeasure is to build institutions that value what they cannot measure — judgment, care, trust, and the willingness to be surprised — and to accept the irreducible uncertainty that comes with it.