What is Goodhart's law?

Goodhart's law says that once a number is used as the goal people are judged on, they change their behavior to improve the number, and it no longer reflects what it used to. It matters because dashboards and bonuses depend on measures that the process of managing them can spoil.

How it works

A measure is useful because it correlates with something we care about. When it becomes a target, people look for the cheapest way to raise it, and those ways often break the link to the real goal. Teachers teach to a test, a team closes tickets quickly without solving the problem, and a seller books revenue early.

The measure improves while the thing it stood for stays flat or gets worse. Defenses include using several measures that are hard to improve together, rotating or hiding the exact measures, and checking the outcome the measure was meant to represent from time to time.

Example

Say a support team is judged on the average time to close a ticket, with a goal of 2 hours. The true goal is that customers get their problems solved. Agents learn that closing a ticket and having the customer reopen it counts as two tickets, each fast. The average closing time falls to 1 hour, but 30 percent of tickets reopen, so the customer's real wait is longer. The figures are an illustration.

How to model it in Machinations

Create a Source for incoming tickets, a Pool for open tickets, and a Gate that decides whether a closure is real or a quick closure that reopens, with the probability set by a Register that rises with a pressure Lever on the closing-time target. A Register shows average close time and another shows the true solved rate. Play with the pressure low and high, and chart both Registers to see them move apart.