Why would the author of a CSS reset or component library wrap its base selectors in :where(), and what does that choice cost?

## The problem base styles have A reset, a normalizing layer, or a component library's default styling exists to be replaced. The author does not know what the consumer will write, only that whatever it is should win. Specificity works against that goal, because the selectors that make base styles *correct* are often the ones that make them *heavy*. Consider a typography base that has to scope itself so it does not leak: ```css .prose ul li { margin-block: 0; } /* (0,2,1) */ ``` A consumer writing `.menu-item { margin-block: 0.5rem }` scores `(0,1,0)` and loses. Their options are all bad: lengthen their selector to outweigh the library, add `!important`, or reach for an ID. Each of those raises the floor for the *next* override, and a codebase where every override escalates ends up with selectors nobody dares to touch. ## What :where() changes `:where()` contributes zero specificity regardless of its contents. Wrap the entire selector and the rule drops to the floor: ```css :where(.prose ul li) { margin-block: 0; } /* (0,0,0) */ ``` The matching behaviour is unchanged — the same list items are targeted — but any authored declaration now beats it. `li { margin-block: 0.5rem }` at `(0,0,1)` wins. A bare type selector defeating a three-part scoped selector is exactly the inversion the library author wants. Note the wrapping must cover everything you want zeroed. `:where(.prose) ul li` still scores `(0,0,2)` because the two type selectors sit outside the function; `:where()` zeroes only its own arguments. ## The costs, honestly stated **You give up winning entirely.** There is no such thing as a base rule that loses to intentional overrides but survives accidents. A consumer's unrelated `ul { padding: 2rem }` will clobber a carefully considered default with no warning and no diagnostic — the styles simply do not appear, and the debugging session that follows starts from "why is this rule greyed out in devtools". **Ordering becomes the only lever inside the library.** Once every internal rule scores `(0,0,0)`, source order is the sole tiebreak among them. A general rule placed after a specific one silently wins, which is the opposite of the intuition most authors carry from ordinary stylesheets. Libraries that zero everything must be disciplined about file order in a way that ordinary stylesheets are not. **It hides intent from readers.** `:where(.prose ul li)` and `.prose ul li` look almost identical in a diff, and a reviewer who does not know the convention will not register that one of them cannot win. Applying it selectively — some rules wrapped, others not — produces a stylesheet whose cascade behaviour cannot be inferred from reading it. **It does not solve override order across sources.** `:where()` only removes specificity from the comparison. It says nothing about which stylesheet loads first, and it does not protect the library from being loaded *after* the consumer's CSS. ## Where it fits, and where something else does The sweet spot is genuinely bottom-of-the-stack defaults: element normalization, typographic rhythm, and component base styles that exist purely so an unstyled page looks sane. It is also useful for adding *context* for free — `:where(.theme-dark) .button` scores the same `(0,1,0)` as `.button` alone, so a theme qualifier does not start an escalation war. What `:where()` is not is a general answer to override ordering. It gives a rule zero weight; it does not give you a way to say "this whole group of styles should lose to that whole group". Cascade layers exist for that grouping problem, and the two mechanisms are complementary rather than substitutes. ## What a strong answer sounds like A candidate who has actually shipped this says three things: the mechanism (zero specificity by definition, matching unchanged), the payoff (consumers override with a plain class and never need `!important`), and the honest cost (the library can no longer defend any default, and internal source order becomes load-bearing). Saying only the first two is the tell that the technique was read about rather than lived with.