# Interface-state evidence study methodology

Version: 2026-09-16.1

The study uses twelve fictional interface-state fixtures on one public SaaS Mirror page. Each fixture declares a visible message, a fact established by the fixture, a fact that remains unknown, a semantic live-region role, an optional next action, and whether sample input remains visible.

The browser run records the actual rendered fixture identifier, message, semantic role, action label, preserved-input value, selected-control focus, CSS viewport and device pixel ratio. The in-app browser's observation sandbox did not expose its user-agent string, so the public record marks that field unavailable instead of inferring a browser version. The run covers every fixture at documented desktop and narrow viewport widths, with the initial loading fixture reset after the other variants. Four representative states receive screenshots from the same built page.

An observation passes when the rendered page matches the reviewed fixture definition. This verifies the public teaching fixture and evidence format. It does not establish behavior in another product, browser family, assistive technology, network service or production environment.

The fixtures contain no customers, participants, private accounts or real product outcomes. They are designed to demonstrate bounded status language. Status-region inspection is not a screen-reader test or an accessibility conformance evaluation.

The public JSON and CSV contain 24 recorded observations: twelve fixtures at each of two CSS viewports. The four screenshot hashes are included in the JSON record. The blank worksheet supports independent repetitions. Corrections should preserve the earlier record and describe what changed.