Search surface

A record should include the product, URL or screen, result type, query, language, region, device, time, and enabled features. These fields are not bureaucracy: without them, two observations may not be comparable. An answer snapshot and its links are more useful than a recollection.

The surface describes where the result appeared; a source is the document or domain shown there. A page can be crawlable without appearing to a user, while a card can show a link without proving that its text was retrieved from that page. Store these events separately.

For comparison, use a fixed query set and repeat it on the same surface. Mark experimental modes, personalization, and account state separately. An interface change can alter link counts and answer shape by itself, so a trend without a change log can be mistaken for a content improvement.

Run ten questions on two surfaces and save complete answers. Code a mention, citation, link, and classic-result position separately. If a domain appears in only one mode, record the result as surface-specific rather than universal visibility.

A search surface does not reveal the system’s internal architecture or establish causality. It says where and when a result appeared, not which index, model, or signal produced it. This matters especially when comparing closed AI interfaces. A surface may be visible while retrieval remains hidden, so its label is not a map of the internal system.