agentleFS
The-Artificer-of-Ciphers-LLCThe-Artificer-of-Ciphers-LLC/skills-from-the-artificer/goodharts-law/SKILL.md

goodharts-law

Skill6 starsChanged 34 days agoMITView on GitHub

Apply Goodhart's Law when designing metrics, OKRs, KPIs, performance reviews, engineering productivity measures, or any system where people are evaluated by a number. Trigger on phrases like "how should we measure this?", "our metrics aren't capturing what we want", "people are gaming the metric", "velocity doesn't feel like a real measure of progress", or any situation where a measurement has become a target. Goodhart's Law is critical for anyone designing incentive systems or interpreting data.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask: does this metric still represent what we want?

    line 41
  2. Distinctive:

    If people understand *why* you're measuring something, they're less likely to optimize the measure at the expense of the goal.

    line 47
  3. Distinctive:

    If someone wanted to make this metric look good without improving what it represents, how would they do it?

    line 54

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Goodhart's Law

"When a measure becomes a target, it ceases to be a good measure." — Charles Goodhart, 1975 (popularized by Marilyn Strathern)

The core idea

The moment you tell people they'll be evaluated by a number, they start optimizing for that number — often in ways that are disconnected from, or actively harmful to, the underlying goal the number was meant to represent.

This isn't because people are bad. It's because optimizing for a proxy is easier and more legible than optimizing for the underlying reality.

Classic examples in software

What you measureWhat gets gamed
Lines of code writtenVerbose, redundant code
Bugs closedBugs closed without fixing, or marked as duplicates
Velocity (story points)Points inflated per story
Test coverage %Trivial tests that don't assert anything meaningful
PR countMany tiny, low-value PRs
Support tickets resolvedQuick closes without real resolution
Time to first responseAuto-responses that count as "responses"

Why it happens

A measurement is a proxy for the thing you actually care about. When you turn the proxy into the goal, you create pressure to optimize the proxy — and the path of least resistance is often to do that without actually improving the underlying thing.

A good engineering team and a mediocre one can produce the same velocity number. A resolved ticket doesn't mean a happy customer. The number gets decoupled from reality.

How to defend against it

Use multiple metrics together, not any one metric in isolation. Gaming one metric is easy. Gaming five simultaneously, when they pull in different directions, is much harder. Pair velocity with defect rates. Pair ticket resolution with customer satisfaction.

Audit the proxy-to-reality relationship regularly. Ask: does this metric still represent what we want? Talk to people. Watch their behavior. A metric that's being gamed will show up as suspicious patterns.

Weight outcomes over outputs. Outputs (PRs merged, tickets closed) are easy to count. Outcomes (users retained, errors reduced, feature adoption) are harder to fake. Wherever possible, measure outcomes.

Make the goal transparent, not just the metric. If people understand why you're measuring something, they're less likely to optimize the measure at the expense of the goal. "We measure velocity to understand predictability, not to evaluate individual performance" changes how people relate to the number.

Don't use a metric for evaluation if it's also used for improvement. Measurement used purely for learning stays honest. Measurement tied to compensation or promotion gets gamed.

Key questions to surface

  1. If someone wanted to make this metric look good without improving what it represents, how would they do it?
  2. Are we seeing suspicious patterns in this data that might indicate gaming?
  3. What are the leading indicators (harder to fake) versus lagging indicators (easy to manipulate)?
  4. Is this metric driving the behavior we actually want, or just the behavior that looks like what we want?

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.