Design an end-to-end MLOps platform on AWS for a regulated lender under SR 11-7.

The reflex that fails this is bolting governance on at the end as an 'audit dashboard'; the senior move is making every control a gate the pipeline cannot skip, so evidence is produced by construction. The held-back follow-up is almost always 'show me how a declined application traces back to its training data', and if your design can't answer it in one query path you've described a research platform, not a regulated one. Have a clear position on segregation of duties: the person who trains a model must not be the identity that promotes it, and that's an IAM boundary, not a policy memo.