StratA11y-1800 Dataset: A Stratified Web Accessibility Research Dataset of Rendered Home Pages
Description
StratA11y-1800 is a research dataset of 1,800 rendered home pages acquired on 24–26 September 2026 from the permanent Tranco ranking L5PV4. It combines reusable page snapshots, automated accessibility measurements and collection provenance to support accessibility research and remediation experiments without repeating the initial acquisition and evaluation. The sample covers five popularity strata within a frame of one million ranked domains: ranks 1–1,000 (200 pages), 1,001–10,000 (250), 10,001–100,000 (300), 100,001–500,000 (450) and 500,001–1,000,000 (600). Each observation represents one captured home page, not an entire website. The deposit contains five files: 1. README.md: An introduction to the dataset, its organization, sampling design, measurements, interpretation and offline analysis commands. Start with this file. 2. StratA11y-1800_v1_draft_2026-10-01.zip: The complete dataset package. It contains the processed table, metadata, documentation, figures and reproduction code, together with 9,000 raw evidence files. Every observation has five artifacts: an original HTTP response document, a rendered DOM snapshot, a screenshot, an Axe report and a Lighthouse report. Extract this archive to access the full evidence and run the supplied analysis scripts. 3. StratA11y-1800_v1_draft_2026-10-01.zip.sha256: The SHA-256 checksum of the ZIP archive, provided to verify that the downloaded archive is intact. 4. data_dictionary.csv: Definitions, data types and groups for the 69 fields in the processed observation table. 5. observations.csv: The processed table with 1,800 observations and 69 fields. It includes public identifiers, URLs, Tranco ranks and strata, design weights, declared language, site categories, DOM characteristics and accessibility measurements. Public identifiers link each row to its five evidence files in the ZIP. The separately uploaded README and CSV files are identical to their counterparts inside the ZIP. They provide direct access without downloading the full archive and do not represent additional observations. Measurements include WCAG-tagged Axe violating-node counts and Lighthouse accessibility scores on a 0–100 scale. Acquisition used bounded scrolling and stabilization to expose common lazy-loaded content. Site-category annotations include 1,799 GPT-6 Luna classifications based on URL/title information and one manually reviewed label; they are not validated ground truth. Unweighted summaries describe the released pages. Design-weighted analyses use the documented sampling assumptions and should account for technical nonresponse and category uncertainty. Offline verification and analysis do not require API credentials. Third-party captured content retains its original rights, as explained in the package's rights notice.
Files
Institutions
- Universidad VeracruzanaVeracruz, Xalapa