I. Goodhart's Original Statement

Charles Goodhart was a British economist, at the time an adviser at the Bank of England. In 1975, in a paper titled Problems of Monetary Management: The U.K. Experience, presented at the Reserve Bank of Australia, he observed something strange: every time the Bank of England picked a monetary aggregate (say, M3) as a target for controlling inflation, that aggregate quickly lost the indicator value it used to have.

Context: in the 1970s, Britain was suffering high inflation. The monetary authority, under the influence of monetarist thinking, decided to target M3 (a broad money-supply aggregate). The reasoning: historically, M3 had a fairly stable correlation with nominal GDP. Control M3, and you'd control inflation.

But banks and financial institutions reacted. They invented new deposit products (not counted in M3), reclassified assets, and kept expanding credit without M3 rising. The target was met on paper - M3 stayed within its band - but the real economy wasn't actually being reined in. M3 stopped being a good measure of economic activity the moment it became a policy target.

Two decades later, in 1997, anthropologist Marilyn Strathern condensed this principle into the line everyone quotes today:

II. Why Does This Happen?

The underlying mechanism is the gap between what we actually care about and what we can measure. We care about the health of the economy - we measure GDP. We care about developer productivity - we measure commits. We care about teaching quality - we measure test scores. In every case, what we measure is only a proxy (a stand-in variable) for the real thing.

A good proxy is one that correlates strongly with the real thing - as long as no one is thinking about gaming it. But the moment the proxy is declared the target - rewards, punishments, promotions, and budget all tied to it - people inside the system learn to maximize the proxy without maximizing the real thing. The correlation between proxy and reality shatters.

Worth emphasizing: Goodhart does not require anyone to cheat. Even if 100% of the actors in a system are perfectly honest, the mere fact that they optimize the proxy (because that's what gets rewarded) means the real thing gets sacrificed along whatever dimensions aren't being measured. This is called the "strong form" of Goodhart's Law.

III. Three Classic Examples

IV. Case 1 - Measuring Software Engineer Productivity

More than half a century after Goodhart, the software industry has replayed the Bank of England's exact trap - the only difference is that M3 has been swapped for "lines of code," "commits," "PRs merged," and, most recently, "AI tokens spent."

These three cases illustrate the same pattern: every popular productivity metric is a proxy, and Goodhart kicks in the moment an incentive is attached to it. The industry's answer - packaged into frameworks like DORA, SPACE, and DX Core 4 - isn't "pick a better metric" but "use several metrics at once, balancing speed against quality, quantitative against qualitative." So that every speed metric is offset by a quality metric. So that no one can optimize along one dimension without it showing up on another.

V. Case 2 - Measuring Vietnam's Pace of Development Through GDP

Moving from nail factories and AI tokens to a national-scale proxy: GDP. And the question is: when a country turns the GDP growth figure into a political target tied to the Party Congress's five-year cycle, what happens?

This is precisely the situation Goodhart warned about: a statistical quantity (GDP) that used to be a fairly reliable indicator of economic welfare is now under pressure for control purposes - specifically, controlling the political narrative of a five-year cycle. The real risk isn't "whether 10% gets hit or not," but the distortions to the statistical and decision-making system once every level of government knows it has to hit the number.

A Lesson From the Neighbor's Backyard

The closest institutional parallel: China. For decades, provincial GDP growth was the KPI that determined the career advancement of provincial officials. Academic papers (including Ni & Jiang 2024, and Diminishing Target-based GDP Manipulation in the Journal of Public Economics, 2025) provide empirical evidence: provincial officials manipulated data to hit their targets.

In 2015, several northeastern Chinese provinces admitted to fabricating GDP data; in 2017, two more provinces were caught. The mechanism: provincial officials were rewarded with promotion based on local growth rates; they reacted exactly per the Goodhart formula - optimizing the proxy (the reported number), not the real thing (their province's actual economic welfare).

Vietnam is not China. The statistical apparatus, the processes, and the degree of independence of the General Statistics Office differ substantially. But the incentive mechanism - a growth KPI tied to the political cycle, quotas handed down to each province - is structurally similar from a game-theory standpoint. The OECD's Economic Survey of Vietnam 2025 and the IMF's Article IV report both stress the same message, in diplomatic language: growth quality matters more than growth speed; the focus needs to shift from inputs (capital) to productivity; the statistical framework needs stronger independence and transparency.

Goodhart Looks at GDP: Three Faces of Distortion

Not all of these faces are bound to occur in Vietnam. But they mark the risk space that Goodhart warns will open up once an economic proxy is turned into a rigid political target. What East Asia Forum calls "a double-digit target that is mathematically unrealistic yet politically useful" is itself an acknowledgment of this point: a high target can force out structural reform if it's interpreted correctly; or force out inflated numbers if it's interpreted the wrong way.

The Question International Organizations Are Asking

OECD Economic Survey Vietnam 2025 and IMF Article IV 2025 propose supplementary indicators alongside GDP - not replacements. The logic is precisely the Goodhart-resistant logic: keep GDP as one dimension, but never let it stand alone.

This set of indicators paints a multidimensional picture that a single GDP figure can't replace. And crucially: because of the trade-off structure between the dimensions (growing fast via credit widens the credit gap; growing via public investment raises public debt) - it's very hard to game all of them at once. This is the Goodhart-resistant mechanism built into macro design.

VI. Conclusion: Measurement Epistemology

Goodhart isn't saying measurement is useless. He's saying measurement has regimes: some uses make for a good measure, others don't.

Three principles for avoiding the trap - equally true for measuring developer productivity and for measuring a nation's pace of development:

1. Use several metrics at once, and don't collapse them into a single number. Keep the dimensions separate so the trade-offs reveal themselves. DORA has 4 metrics; DX Core 4 has 4 metrics; the IMF/OECD recommend viewing GDP alongside TFP, ICOR, credit gap, and debt service.
2. Balance speed metrics against quality metrics, quantitative against qualitative. Every number that's easy to inflate must be paired with one that's hard to fake. In engineering: deploy frequency paired with change failure rate. In macro: GDP paired with job quality, regional disparity, and budget sustainability.
3. Separate diagnostics from targets. Not every metric needs to become a KPI. Many metrics are useful precisely because they aren't tied to an incentive. Turning them into a KPI destroys that usefulness - exactly per Goodhart's original 1975 statement.

From the Bank of England in 1975 to Microsoft in 2026 and Vietnam's 2026–2030 GDP target, the principle hasn't changed. Systems of people are smarter than systems of metrics. If a metric is single-dimensional and carries an incentive, people will find a way to make the number look good without making the real thing any better - usually not out of malice, but because that's the rational response of an actor facing a poorly designed system.

Goodhart's Law doesn't tell us to stop measuring. It tells us to measure more, across more dimensions, and with more humility about what any single number can actually tell us.