Flaky Tests — 15 Causes and Proven Fixes (2026)

Last updated 2026-07-20 · 13 min read · By Avinash K

Flake rate is the single strongest predictor of whether a QA team ships on time. Above 5% flake, engineers stop trusting CI; above 10%, they stop reading it. Here are the 15 causes we see repeatedly across audits — and the proven fix for each.

Key takeaways

• The 15 root causes grouped by category.
• Playwright, Cypress, and Selenium fixes for each.
• A weekly triage ritual that keeps flake below 2%.
• The one metric to track (Flake Rate, not Pass Rate).

1. Fixed sleeps — replace with auto-waiting locators.
2. Race on network — page.waitForResponse then assert.
3. Animation not settled — disable animations in test config.
4. Focus stolen by another tab — pin browser tab, disable notifications.
5. System clock skew — freeze time with fake timers.

6. Shared test data — each test creates and cleans its own user.
7. Global state — reset localStorage / cookies between tests.
8. Parallel worker collision — namespace by workerIndex.

9. Flaky third-party — stub in tests, contract-test separately.
10. DB seeded once, mutated by tests — seed per test.
11. External auth provider throttling — use a test tenant with high limits.

12. CPU-starved CI runner — raise runner size; do not add retries.
13. Chrome version drift — pin browser version in Playwright config.
14. Docker DNS caching — restart containers between suites.
15. Retry masking real bugs — cap retries at 1 and quarantine on second failure.

// Bad
await page.waitForTimeout(3000);
await page.click('#submit');

// Good
await page.getByRole('button', { name: 'Submit' }).click();
await expect(page.getByRole('alert')).toBeVisible();

See the Playwright masterclass and Selenium interview questions for framework-specific patterns.

1. Every Monday, list top 10 tests by flake rate.
2. Owner assigned within 24h.
3. Quarantine tests above 20% flake — do not run on PRs.
4. Hard cap: quarantine expires after 5 days; test is fixed or deleted.

Reference: Google Testing Blog on flakiness.

Most of the flake we trace back to design, not infrastructure: tests written after the code, against selectors and states nobody planned for. If that sounds familiar, work through our test-driven development (TDD) guide for QA engineers — writing the assertion first removes an entire class of timing and isolation flake before it reaches CI.