Lambda Land

Goodhart’s Law is invoked to argue that once a metric becomes a target, it stops reliably indicating underlying performance; the author refines this into a corollary: only collect metrics you’re willing to turn into explicit targets. Measurement reshapes incentives because organizations act on the information they have, so changing what is measured changes behavior. This is presented as a practical steering principle: altering an organization’s access to information will alter what it prioritizes, and avoiding harmful targets requires discipline about what data you surface.

A concrete workplace anecdote illustrates the risk: leadership celebrated a year-over-year 2× increase in commits, 5× more pull requests, and 5× more lines changed. The author challenged these celebrations, noting that lines-of-code and crude volume metrics reward churn and erode code quality, and questioned whether many more PRs implied much less comprehension of the codebase. Leadership later framed the numbers as AI experimentation, but the core lesson remains - easy-to-get GitHub stats invite harmful targets, so the safest guardrail is not collecting or publicizing metrics you don’t want turned into goals. Acknowledgement is given to Mike Munger for the idea.