Structure is programmatically determinable — semantic headings, lists, tables, and form groups; headings and labels describe their content

Requirement

Express the page's structure in markup, not just visually: real heading elements in a logical hierarchy, lists as ul/ol, tabular data as tables, related form fields grouped. Headings and labels describe the content they introduce; link text makes sense read on its own.

Rationale

The accessibility checklist Essential tier ("Presentation of info and relationships" + "Headings and labels" + "Link text and purpose"). A person using a screen reader navigates by headings and landmarks — visual-only structure is invisible to them. This is also what makes pages legible to agents and checks: semantic markup is machine-readable structure.

Passes when

• One h1 per view; heading levels nest without skipping; sections that look like sections are sectioned.
• A roster rendered as a list IS a list (ol/ul); data grids are tables with header cells.
• Headings name their content ("Attendance", "No notes yet"), not generic ("Overview", "Section 2").
• Link text describes the destination ("Make a correction"), never "click here".

Fails when

• Styled div/span standing in for headings, lists, or buttons.
• Heading levels chosen for font size rather than hierarchy.
• A label that only makes sense with visual context ("More", "Here", bare icons).

How to verify

Static half: checks/structure-scan.py (the structure check) flags a <table> or a role="table" element whose subtree holds no <th> and no role="columnheader". That one table rule is the whole static half. It runs the same walk for CMP-6 and reports the finding under both ids.

Rendered half: the rendered runner's axe rules heading-order, list and listitem decide heading order and list semantics. No static heading walk ships: a source scan cannot see a page assembled from a layout, MDX and components, so it would flag good work.

Judgment half: the evaluator reads headings, labels and links out of visual context and judges whether they still communicate, and decides whether styled divs should have been a heading, a list or a table. Form grouping (fieldset / legend) stays manual: it needs to know which fields are related.