All 200+ tests passed, but the UI was completely broken

There is nothing worse than pushing a clean CSS refactor with zero failing integration tests, only to realize a shared utility class broke three nested components on a screen you didn't check manually.

Relying purely on functional assertions leaves a massive blind spot for visual bugs.

We recently set up a basic script to capture and diff pixel level screenshots on mocked routes before shipping changes.

How do you guard against accidental styling regressions on solo or small team projects?