Faceted search

Faceted search splits a set of results along several attributes at once. Each attribute is a facet (type, availability, price band, brand), and each facet is its own group of values. Values inside one group usually widen the result set, because ticking two of them means either will do; groups combine the other way, because a reader who picks Lamps and In stock wants both to be true. That pair of rules is the whole logic of the pattern, and it is why a facet rail can narrow thousands of items in three clicks without anyone writing a query.

The counts are not decoration. A facet value labelled with the number of results it would leave turns filtering into a series of informed moves rather than a series of guesses, and it is what keeps a reader out of a dead end: a value that would return nothing can be counted zero, dimmed, or dropped, so no one spends a click reaching an empty page. Counts are also expensive to compute honestly, which is why they are the first thing dropped by an implementation in a hurry, and why a rail of plain checkboxes is a weaker version of the same idea.

Applied filters have to be visible and individually removable somewhere the reader is already looking, usually as a row of chips above the results. Without that, the state of the query lives only in a rail that may be scrolled out of view or collapsed behind a button on a narrow screen, and an unexpectedly short list reads as missing stock instead of as a filter still switched on. Removal should be per value, not just “clear all”, since the usual repair is loosening one constraint rather than starting over.

The pattern fails on data before it fails on interface. Facets are only as good as the metadata behind them, so a catalog where half the items have no material recorded will show a Material facet that hides more than it reveals. Too many facets is the other failure: past roughly six or seven groups the rail becomes a form, and a reader who has to read it is no longer browsing. Search answers a question with words; faceted search answers it by removing everything the reader said they did not want.