The short answer
Provide a meaningful caption, explicit row/column headers and complete units and missing-value rules. Use native data-table structure rather than a screenshot or visual alignment alone. Keep a wide table in a named keyboard-scrollable region and recheck exported formats separately. A readable grid does not prove that a PDF, copied image or destination retained its structure.
What you’ll take away
- Keep semantic header relationships as well as visible labels.
- Explain observed zero, unknown and the exact percentage base.
- Check the actual export and task, not just its appearance.
Define the complete reading task and row unit
A useful recap table lets a reader identify the period, eligible unit, measured count and denominator without an author standing beside them. The data dictionary explains field meanings; the table itself must still communicate its title, header relationships and missingness at the place it is read. Do not make essential context available only in a private source file.
Use one coherent question per table. If a grid combines unrelated populations or changes the denominator mid-column, simplify it before adding styling. A table is not automatically accessible because its text is selectable, and a neatly aligned paragraph does not acquire row/column relationships merely by looking like a grid.
Read the complete constructed outcome table
Constructed period A has 9 successes among 12 classified outcomes, plus 3 eligible outcomes still unknown. Its known-outcome rate is 9/12 = 75%. Period B has 4 successes among 8 classified outcomes and 0 unknown: 50%. The unknown column is separate; A is not a measured 75% success rate for all 15 eligible units.
The supplied HTML gives the table a descriptive caption, column headers and period row headers. A visible note explains that the counts are fictional, zeros are entered only when known, and the displayed percentage uses classified outcomes. The complete source can be read or adapted without uploading private data or installing a library.
| Period | Successes | Classified | Unknown | Rate among classified |
|---|---|---|---|---|
| A | 9 | 12 | 3 | 75 % |
| B | 4 | 8 | 0 | 50 % |
Preserve the relationships in the actual format
For a simple HTML table with headers in its top row and first column, W3C demonstrates th cells with scope="col" and scope="row". Use td for data and caption for the table’s subject. This is a concrete structure for the supplied example, not a reason to add an ARIA grid or put every data cell in the Tab sequence.
More complicated grouped or multi-level headers need the appropriate associations, potentially including scope groups or id/headers. A simpler set of separate labeled tables can be easier to maintain. W3C warns that conversion can lose table markup; a PDF needs its own structural and reading review rather than inheriting an HTML pass.
Resolve three constructed table handoffs
T1 — Constructed outcome table: A shows 9, 12, 3 and 75%; B shows 4, 8, 0 and 50%. Decision: label the known-outcome denominator and unknown column explicitly. Outcome: do not convert A’s unknown results to failures or call 75% the verified whole-cohort rate.
T2 — Constructed screenshot handoff: the original has proper table headers, but the delivered artifact is only an image. Decision: retain a real readable table/text equivalent and inspect the destination. Outcome: visual similarity and extracted characters do not establish preserved header relationships.
T3 — Constructed multi-level grid: several grouped populations share ambiguous blank header cells. Decision: simplify into separate titled tables or implement and inspect the actual required associations. Outcome: no structural certificate from bold styling, merged cells or an automatic checker alone.
Keep narrow reading and keyboard exit usable
A genuinely two-dimensional table may need horizontal scrolling. Keep the surrounding explanation readable, name the scroll region and make it reachable without a pointer. In the supplied example, the region can receive keyboard focus; use the browser’s scrolling controls and then Tab or Shift+Tab to leave. Native table cells remain native cells.
Record the actual browser, width and task when checking this path. A keyboard scroll check is not a screen-reader study or proof about every device. Avoid shrinking essential table text into an unreadable image just to remove a scrollbar. Keep a complete usable alternative when the intended publishing surface cannot preserve the needed structure.
Keep data, markup and export checks together
The worksheet keeps question, caption, row unit, header associations, units, zero/unknown rules, exact source values, delivered file and actual keyboard/reading result. The packet includes full bilingual semantic HTML and the constructed data. Changing values requires rechecking both the source and the visible explanation.
W3C table and keyboard guidance was read on 1 November 2026. The example is original constructed material, not a real reader cohort or proof of complete accessibility. No private dataset, assistive-technology participant or native account was used to create it.
Sources & product references
Product details reflect the publication date. External features and interfaces may change.
