App Crash Lab

A web app can say Saved while the next reload brings back an old value. App Crash Lab focuses on that gap. You give it a working local save flow, the values that should persist and, when applicable, an invalid update that the app should reject. It then compares the saved state after those actions.

The output includes baseline and observed values, screenshots, traces and a standalone Playwright test. Keeping that test is the useful part for a developer fixing a regression: the same expectation can be checked again after the change.

The public browser lab illustrates a weak assertion with a deliberately broken language example. Counting elements labelled English can miss a message that has changed to Japanese. Adding the expected message text gives the test a more specific condition. This is an authored demonstration, not a newly discovered production bug.

The CLI has the same dependence on a well-defined contract. A URL alone does not tell it which values are correct. The documentation distinguishes an inconclusive setup or baseline from a failed preservation check, which helps avoid treating a test-configuration problem as an app defect.

The current public page identifies version 0.3.1 and an MIT license. The CLI needs Node 22 or newer and Playwright with Chromium. The project says the runner needs no account, API key or runtime model call. Initial dependency and browser downloads still need internet access.

Use a disposable local app and synthetic records for these checks. They cover the selected values and explicit expectations, not every failure mode; power loss, multi-user races and service workers are outside the stated scope. This note draws on the public documentation and example pages. The CLI was not installed or run.