Methodology

The unit of evidence on this site is a specific source or a specific executed observation, not an overall impression that a component works.

Sources

Documentation records identify the primary page, its version or visible update date, and the date it was opened. If a source describes a capability, we say it is documented. Missing information is not turned into a negative result, and a documented feature is not turned into a runtime pass.

Observations

Executed records identify the package or original test page, the exact steps, the browser name and version, the operating system, the date and the observed state. We record what can change a result: content, dimensions, input method and relevant settings. A failed or blocked check is kept with its limitation rather than replaced with an assumption. When a fix changes a result, the earlier run stays on record.

Keyboard operation does not establish screen-reader announcements. Programmatic or emulated scrolling does not establish physical touch behaviour. A narrow container does not establish every narrow screen. Results apply to their recorded environment and test page.

Synthetic data

Chart and table examples use labelled synthetic data to expose declared cases, and distinguish missing observations from numeric zero. A value-equivalence check compares stated outputs; it does not establish source-data accuracy or assistive-technology behaviour unless those were tested separately.

Datasets

Dataset observations use four statuses: yes, no, partial and not tested, each with a short note. Unknown stays unknown. Every row records the package version, source, licence, file hashes and configuration, so it can be reproduced. Filters help you inspect the evidence; there are no scores, no rankings and no paid rows. Observation data is licensed under CC BY 4.0.

Retesting

We retest after any change that affects a claim and keep the earlier result alongside the correction. Reviewed pages show who repeated the checks and when.