The law Goodhart did not write

Everybody in this industry can quote Goodhart’s law.

When a measure becomes a target, it ceases to be a good measure.

It is on slides. It is in retrospectives. It gets deployed the moment somebody proposes a metric, usually as a way of ending the conversation.

Charles Goodhart did not write it.

What he wrote, in a 1975 paper on British monetary policy, was this:

Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.

The popular version came from an anthropologist. Marilyn Strathern, writing in 1997 about accountability in higher education, gave the idea the phrasing it now travels under, citing a formulation from Keith Hoskins the year before. Her sentence is better. It is shorter, it scans, and you can say it out loud without a second run-up.

It is also a different claim.

What got lost in the compression

Goodhart was an adviser to the Bank of England. The Bank had noticed stable relationships between certain measures of the money supply and inflation. It began targeting those measures. The relationships broke down.

His claim is mechanical and specific: the regularity collapses. Not the measurement. Not the honesty of the people being measured. The statistical relationship between the indicator and the thing you actually cared about. That is what stops holding, because the pressure you applied is now one of the forces acting on the indicator.

Strathern’s version relocates the failure into the measure itself. It “ceases to be a good measure,” which sounds like a property of the measure and quietly invites the conclusion that the measure was badly chosen. Pick a better one and carry on.

That is the reading I hear most often, and it is the expensive one, because it suggests the problem is solvable by choosing more cleverly.

Why the distinction has teeth

Take a security team measured on mean time to patch.

Under the popular reading, the failure mode is gaming: somebody starts closing tickets early, or reclassifying findings, and the number stops describing reality. That happens, and it is worth watching for. But it frames the problem as a behavioural one, with a behavioural fix, better definitions, tighter audit, more scrutiny of the people producing the number.

Under Goodhart’s actual claim, something worse is going on and nobody has to behave badly for it to happen. Mean time to patch correlated with security because, in an unpressured organisation, teams that patch quickly tend to be teams with good asset inventory, working change management, and slack in the schedule. The number was a symptom of those things. Apply pressure directly to the symptom and you get faster patching bought out of the same finite budget that funded the inventory work. And the correlation that made the metric worth watching quietly stops holding.

Everyone involved is honest. Everyone hits the target. The number no longer tells you what it used to tell you, and there is no better number sitting behind it, because the property that made this number informative was that nobody was pushing on it.

What to do with that

I do not think this means stop measuring. An organisation that measures nothing is not thereby well run; it is just unobserved.

It means something narrower. A metric is a claim about a relationship, and the relationship is a thing you have to keep re-earning evidence for. The useful habit is not choosing better metrics up front. It is going back, once the metric has had pressure on it for a year, and checking whether the correlation that justified it in the first place is still there.

Almost nobody does this. We treat the choice of metric as the hard part and the maintenance as bookkeeping. Goodhart’s actual sentence says the opposite: the choice was never the hard part, because the collapse is not caused by choosing wrongly. It is caused by pushing.

Strathern’s line is the better sentence. Goodhart’s is the better warning.