Raw data format

This page describes the raw recordings the reporter keeps in ui-coverage-report/data/, which the report CLI rebuilds the report from. Most users never read these files; they matter if you archive runs or write your own tooling.

The raw data is input for rebuilding the report, not a stable contract like coverage.json. Its types are exported (StoredData, StoredAttempt, RawCoverageData, …) and versioned with formatVersion.

The data folder

ui-coverage-report/
  index.html, coverage.json, screens/   ← the report
  data/
    ui-coverage-data.json               ← every test attempt's raw recording
    screenshots/                        ← every screenshot, not just the ones the report shows

The reporter writes it unless saveData: false.

ui-coverage-data.json

ui-coverage-data.json has formatVersion: 1 and an attempts array: one entry per test attempt and browser context with testId, retry, passed, title, project, trace, tags, specFile and data, the recording (formatVersion: 4, see RawCoverageData in src/model/raw.ts; versions 2 and 3 are still read, as recordings without steps):

• nodes: a table of DOM elements, each { tag, parent?, nth?, of?, attrs? }: the parent's index, the position among same-tag siblings (:nth-of-type), and the attributes. Ancestor chains are stored once and shared by every element under them. Tree boundaries are pseudo-nodes: an element at the top of a shadow root has a { tag: '#shadow-root' } parent, whose parent is the shadow host, and a frame's <html> has a { tag: '#document' } parent, whose parent is the <iframe>.
• snapshots: { url, dpr, viewport?, screenshot?, screenshotClip?, elements, t?, hash? }, with each element { node, name?, x, y, w, h, pos? }. name is the accessible name, used for labels only; candidateOnly: true marks elements recorded although they aren't interactive, like <label>; pos is fixed or sticky when the element or an ancestor is. Boxes are in document CSS pixels. viewport is the scroll offset and viewport size; screenshotClip is the document region a viewport screenshot shows. t is when it was taken (ms since the epoch) and hash the SHA-1 of the screenshot file, so the report can tell the same state reached by several tests.
• interactions: { event, url, element: { node, name? } }, pointing into the same node table, so an interaction resolves to the same key as the element in the inventory.
• steps (with the recording option steps: 'on'): { t, event, url, page?, element?, call?, state? } in the order they happened. event is the DOM event or load; page numbers the context's pages in creation order (absent for the first); call is { api, locator?, arg?, file?, line?, column? }, the Playwright call (file relative to the project's testDir); state indexes states.
• states: the viewport screenshots taken after steps, shaped like snapshots (with the elements in view). They don't count for coverage.
• rules: the deprecated use.uiCoverage identity rules, if any.

Attribute values are recorded up to 256 characters (class up to 1024), and style isn't recorded.

The file also has config, the resolved config the reporter used; the CLI uses its profile when no other is chosen.

Test attachments

The same recording is the ui-coverage test attachment (with one ui-coverage-screenshot-<n> attachment per screenshot), so merge-reports and the data folder carry identical data.

Attachments from earlier versions (no formatVersion, keys computed in the browser) are still read, but report-time rules can't change their keys.