Retries and failed tests

This page explains which test attempts count when Playwright retries tests, and how elements touched only by failing tests are shown.

Playwright runs a test again on retry, and every attempt records coverage. The reporter combines the attempts of each test like this:

• Element totals use inventories from every attempt, passed or failed. A page a failing test reached still adds its elements to the total.
• Tested comes from passing attempts only (by default). An attempt passes when its status is the one the test expects: passed, or failed for a test.fail() test. Skipped, timed-out and interrupted attempts don't pass.
• Retries: if any attempt of a test passed, only the last passing attempt's interactions count. For a flaky test (fails, then passes on retry) that is the retry that passed. The earlier failed attempts add element totals only.
• Tests that never passed: their interactions don't count as tested. Elements that only such tests touched are shown as failing tests only (amber) in the report. They count as untested in the percentage. In coverage.json these elements have tested: false and a non-empty failingTestIds, and each view and the summary have a failingOnlyElementsCount.
• No duplicates: a test is listed once under "Tested by", however many attempts it took.

The same "counted test" rule decides whether a link is tested: a counted attempt must have visited the destination view.

Counting failed tests

With countFailedTests: true, interactions from every attempt count as tested, whether it passed or not:

// ui-coverage.config.ts
import { defineUiCoverageConfig } from 'playwright-ui-coverage/config';

export default defineUiCoverageConfig({
  countFailedTests: true, // default: false
});

countFailedTests is applied at report time, so a rebuild can switch it.

Links to the HTML report

When "Tested by" links to Playwright's HTML report, a flaky test links to the retry that passed, and a test that never passed to its last attempt. See Insights and the HTML report.