If your AI made a bad call last month, what could you actually show?

Genuine question for anyone running AI in production.

Say a model made a decision — approved something, flagged something, scored someone — and weeks later it gets challenged. A customer disputes it, or a regulator asks, or legal wants the record.

What do you actually have? A log you could've edited? A screenshot? "Trust me"?

I've been building in this space (AURORA — evidence custody for AI decisions) and the thing I keep hearing is that most teams don't find out their records won't hold up until the moment they need them to.

Curious where people actually land:

- Do you keep tamper-evident records of AI decisions today?

- Has a decision ever been challenged after the fact?

- Or is this a problem that just hasn't bitten yet?

Not looking to pitch — genuinely want to know if this is felt or still theoretical.