I know some have been real bugs. I don't know much "some" is.

sometimes that’s enough to be a bug unfortunately 😅 sometimes my code has broken because I spelled colour or authorisation the British way instead of spelling it the American way.

A very small minority are actually real. It's just fuzzing with extra steps, which is in lay terms the practice of rigging up a test environment that just sends random inputs (sometimes intentionally made to be extra adversarial). But every problem you find by fuzzing is real because there's not an alphabet soup ladle in the way making up bullshit, either a crash happens or it doesn't and you only need the ability to replay those inputs and info on the state of the system from each failure to get a repeatable thing you can investigate. We're drinking lakes to make ourselves worse at everything we were already good at. Completely pointless.