Start with a reproducible failure
Choose an interior data cell and state its expected row header, column header and unit. Use a screen reader’s table navigation. Compare announced headers with the visual model after sorting and mobile adaptation.
Use a small controlled test to separate a symptom from its cause. Record the starting state, affected URL and observed outcome so that another person can repeat the check.
A fictional website example
Fictional example: A delivery table shows times by region and product. Rows become cards on mobile, but column labels are also hidden from assistive technology. The number three loses its unit and context.
Collect evidence before changing anything
Test first and last rows, an empty cell, a merged header and an exception row. Explain a dash if it means something other than a missing value.
| Check | Expected outcome |
|---|---|
| Starting condition | A table can look correct while losing the relationship between value, row and column for someone who cannot see its layout. Start with the question the visitor needs to answer. |
| Implementation | Choose an interior data cell and state its expected row header, column header and unit. Use a screen reader’s table navigation. Compare announced headers with the visual model after sorting and mobile adaptation. |
| Verification | The visitor identifies the correct value with row, column and unit without guessing from position or color. |
Fix the cause and repeat the test
Repair semantics with genuine table headers and relationships. Simplify or split complex tables if header levels become difficult to follow. Preserve headers when visual presentation changes.
Repeat the same test after the change and add a nearby case that already worked. One successful case establishes that task in those conditions; it does not certify the whole website.