Until someone in QA who keeps running that test flow manually says “Wait a minute, my time is valuable!” and scripts the checkout flow in Selenium...

That's what the dev/staging environments are for. Even if you run it 3x/day in production, it's less than $100/month. Use a company card and you get the money back too.

I don't see enough of a problem that requires all the technical debt to support fake numbers in production.

This is the difference between unit and integration tests.

Manual vs selenium? No it isn't...

I've never had it happen and it shouldn't be a problem for occasional spot checks. Make sure everything in dev/staging environments is working first before doing a production run.