Which user interface standards age well, and which never do

Iguazú Falls seen from a forest overlook, dozens of waterfalls pouring over a green cliff under a cloudy sky, with a small boat in the mist below

I have spent a little over a decade writing, reviewing, and arguing about standards for user interface (UI) code. Some are still in use. Some were obsolete within a year. I assumed the difference was how well they were written. I now think it is what they tried to standardize.

The ones that last describe boundaries

The standards I have watched survive one framework after another say every interactive element works from the keyboard, every image has a text alternative, state flows one direction across a component boundary, a shared component has one owning team, and the design system owns color, spacing, and type.

They describe a boundary, an ownership, or an outcome, and were as true for jQuery as for React. My test is whether a standard could survive a complete change of framework. If so, it describes the software. If not, it describes the framework.

The ones that die name a library

The other kind says use this library, at this version, in this folder layout, with these settings, and this state management.

Many are good decisions when made, but they are decisions, not standards, and the current best answer has a shelf life. The folder layout that made sense at ten components makes no sense at four hundred. Then the standard is either ignored or enforced past its usefulness, which is worse.

Formatting rules, I think, belong in tooling, where a formatter enforces them for free and nobody argues in review.

What a standard is for

I have come to believe a standard exists to make a decision once so a team does not have to make it four hundred times. Decisions that come up rarely, that a tool can enforce, or that will be different in two years anyway are better left as recommendations with a date on them.

A standard should say why

The rule I apply is that a standard has to carry its own reason, in a sentence. “Every component must work from the keyboard, because some of our users cannot use a mouse.” “Shared components have one owner, because changes without an owner break three teams at once.”

If I cannot write the reason, I treat the rule as a preference. Preferences are fine, but should not have the force of a standard, because nobody can tell when a preference has stopped applying. Everybody can tell when a reason has.

Where this stops being true

A small team on a short-lived product does not need most of this. If the thing will be rewritten in a year, only the current answer matters.

At some size of organization, naming the library is right, as a platform decision, so that forty teams do not each pick one. A platform choice with a review date is a decision. A sentence still true after the platform changes is a standard. Confusing them is how a standards document ends up ignored.

Photo source: https://photos.robertstowe.com/iguazu-falls