Goodhart's Law

Comparing two versions of something by randomly showing each to different users and measuring which performs better, a randomized controlled trial for products.

Chasing a game's list of secondary objectives, and the argument over whether that list improves play or replaces it.

A stated number pulls subsequent estimates toward it, regardless of relevance.

Over-trusting an automated recommendation, including against available contrary evidence.

The automatically-derived list of pages linking to the current one.

Measuring how fast code runs under controlled conditions, which is far harder to do honestly than it looks.

A coarse language for how an algorithm's cost grows as its input grows.

The proportion of conditional outcomes exercised — the strictest ordinary coverage metric, and the most informative.

The hypothesis that visible disorder invites serious crime, and the long dispute over whether the evidence supports it.

The button or link that asks a visitor to take the next step, such as signing up or buying, whose wording, prominence, and placement are heavily tested and easily turned manipulative.

Cryptographic attestation of an originating telephone number, and why spoofed calls survive it.

The convention that a conspicuously introduced story element must pay off later.

The proportion of code executed by a test suite — a map of what is untested, not a measure of quality.

a genre of indirect settlement control where production chains and mood systems generate stories rather than just outcomes.

Google's small set of user-centered performance metrics: loading, interactivity, and visual stability.

A build check that fails when coverage falls below a threshold, and what that pressure produces.

A team's shared checklist of what must be true before work counts as finished, such as reviewed, tested, documented, and deployed, which keeps hidden work from piling up behind items marked complete.

How it feels to build software with a given set of tools, codebase, and process, shaped mostly by how fast feedback arrives, how much a developer must hold in mind, and how often they can work uninterrupted.

Crash at the point of misconfiguration rather than degrading quietly into a wrong state.

Synthesizing extra frames between rendered ones, which raises displayed frame rate but not responsiveness.

Zero-price entry monetized after the fact, with design that bends toward whoever is willing to pay.

Selling randomized draws rather than items, and the structures that make the real price hard to see.

Never attribute to malice that which is adequately explained by stupidity.

A design tradition that simulates consistent rules rather than scripting outcomes, so solutions the designer never planned still work.

Making a system countable and administrable, at the cost of the local knowledge that made it work.

A game operated as a running service rather than shipped as a finished artifact, with the consequences that shift implies.

Friction between what a game's mechanics reward and what its story claims to value.

Persistent advancement that survives a failed run, converting loss into currency and a hard genre into a habit.

The layer of play above the written rules — the prevailing consensus about what actually works, and how fast that consensus moves.

A macroeconomic account holding that a currency issuer is constrained by real resources and inflation rather than by revenue.

How software introduces itself to new users, ideally by getting them to a first real success quickly rather than through tours and tips they skip and forget.

Ratings and written opinions that customers post about products, places and services, which have become one of the main inputs to buying decisions and, for the same reason, a constant target for fakes, bribes and coordinated attacks.

A model that fits its training data too well has memorized noise instead of learning the pattern that generalizes.

The unit of an absolute change in a percentage, and the gap between that and a relative change.

A committed numeric ceiling on a page's cost, enforced automatically.

In a hierarchy, people get promoted based on performance in their current role until they reach one they're bad at, and stop there.

Divergent interests and unequal information between whoever delegates and whoever acts.

Shaping a model's input to make the desired behavior reliable rather than occasional.

A family of vitamin-A derivatives that act on nuclear receptors, arranged in a potency ladder that is not a ranking of usefulness.

The energy a rolling tire loses to deformation, and why more pressure stops helping.

The two error rates of a test, and why neither answers the question a person actually asks.

A dice-plus-modifier resolution borrowed from tabletop play to decide whether an attempted action succeeds.

The rate at which tissue absorbs radiofrequency energy, measured in watts per kilogram.

A community-authored ordinal ranking of a game's options, and what the ranking actually measures.

A design in which the scarce resource is the schedule itself, so every activity is paid for with another left undone.

Code no input can execute — usually a guard against a state the surrounding invariants forbid.

A closed loop that moves heat by boiling and condensing a working fluid at two pressures.

Measuring how people use a website, from parsing server logs to JavaScript page tags and Google's free Analytics, with metrics like pageviews and bounce rate that are distorted by bots, blockers, and consent rules and easy to optimize in the wrong direction.