Announcement order is structural order
Aliases: source order · AT traversal order · programmatic reading order
What it is
The next sentence a screen reader speaks is the next node in the structural tree, not the next patch the eye would hit on the canvas. Announcement order is structural order: the sequence of nodes in the source and the accessibility tree is the sequence assistive technology walks. Layout can paint a button in the lower right; if the tree still places it after the footer links, what is heard is “footer, then button.”
This is a criterion on authors, not an account of how a reader caches the page. Get the tree in the right order and announcement can be right; get the tree wrong and no amount of visual flow repairs the sequence that arrives at the ear.
Why it happens
Assistive technology does not consume the paint. It consumes a structural projection. Authorial sequence in the source becomes a walkable sequence in the accessibility tree. CSS changes paint and usually not that tree, so “the next thing you see” and “the next thing you hear” are not checked against the same list by default.
Tasks done by ear depend on cause and effect in the sequence: label then field, explanation then action, product then purchase. Reverse those ends in the structure and the story heard is reversed — not a missing sentence, an inverted causal chain. Keyboard focus often follows structure too, but focus jumping is a different operability problem. What this leaf locks is whether the spoken reading sequence is the sequence the author meant people to understand.
Studying it
Have screen-reader users (or evaluators in NVDA / JAWS / VoiceOver browse mode) walk a piece of content without looking, and write down the object order heard. Have sighted people mark their visual reading order. The first divergence is where structure failed to carry the reading. Independent variable: node sequence in the source (visuals held constant). Dependent: sequence agreement, acting on the wrong object, reports that it “sounds like it is jumping.”
Do not treat “how a virtual buffer is built” as the thing to measure. Measure whether the order authors put in the structure is the task order. Reader/browser combinations are a robustness check; the criterion itself holds across them.
Where it stops holding
Canvases, maps, and game objects were never a linear list; structural order has almost no “correct reading,” and spatial exploration is a different strategy. Graphics that never enter the accessibility tree are out of scope. Tables and lists, once they have matching semantics, walk by row or item rules; do not judge them as a whole-page top-to-bottom stream. When ordinary document flow already matches the task, there is no second order to invent.
Applying it
- Write structure in the order the task happens: heading and explanation before the action, label before the field, product before purchase. Place things visually with layout; do not reverse nodes in the tree to please the picture.
- For each main template, produce a list of “what should be heard, in order,” and compare it with source node order — not with left-to-right on the comp.
- How to check: screens off, browse-mode the primary path, write names in the order heard. Ask someone who has not seen the mock “what is this doing.” If the story does not parse, or runs opposite the steps you give a sighted person, fix sequence in the tree; do not paint another visual arrow.