a11y-perception

What of a page is perceivable: a snapshot, its reading order and the difference between two snapshots. The package computes, it does not collect.

Status

0.2.1. Extracted from auditmysite (accessibility/{tree,diff,snapshot}.rs, screen_reader/{linearizer,types}.rs) after the rework that was measured on 133 pages with 692 journey instances. 27 tests came along and run without a browser.

New in 0.2.0 is the announcement structure (announce). The renderer from auditmysite’s announcer.rs could not be separated from it, because it handled structure and words in one go: which parts an announcement has, in which order, and which state adds nothing — and right next to it, what that is called in German. The first half is the same everywhere and now lives here; the second stays with the host, which brings its own localization.

Structure

Public surface

Entry	Purpose
AXTree, AXNode, AXValue, AXProperty	the tree and its nodes; node_by_backend_id, is_within for containment
AXSnapshot, FocusSnapshot, FocusIndicatorStatus, Rect	one captured point in time including focus position
AXTreeDiff	the difference between two snapshots
linearize, ReadingItem, IgnoredReadingNode	the projection into reading order, with what it skips
announce, Announcement, AnnouncedRole, AnnouncedState	what the announcement of a reading unit consists of — named parts, not a string

Dependencies

Downward: only serde. Upward: auditmysite, Relief. Deliberately not a11y-report: this package produces no findings.

Limits

Three things the measurement taught, which apply here as rules:

1. “Added” means became perceivable. Chrome does not remove a hidden node — it stays with the same backend ID as ignored with reason notRendered. Defined by presence, the diff missed the most common way of opening up content.
2. Identity via the backend ID, not via position. Comparing positions across two arrays was one of the three reasons why 44% of the high findings of a JavaScript probe were wrong.
3. Unclear identity is reported, not guessed. Missing backend ID, node replacement, frame change, navigation.

And the limit of the method itself: the diff does not prove an announcement. Transient changes between two snapshots need an event trace from a host — in the originating project that is interaction::live_regions, and that is collection, so not here.