Every page, opened before every deploy

Type checks and unit tests don't catch a page that throws when real data arrives. So before a deploy we open every page in headless Chrome.

A test visits each page and saves the data it would send. A script serves the built frontend and that saved data from disk, blocks every other host, and opens each page. It fails on an uncaught error, a console error, or a call to our error reporter.

Nothing leaves the machine and nothing needs a login, so it runs anywhere.

Where it falls short

It only knows the data the test gave it. A test stub that doesn't match the real page will pass. We learned that one the hard way.