Agree that serious review too late is very annoying for all involved. Reviewers are unhappy they have to reject, implementers are unhappy to have their work rejected, and product folk are unhappy that their feature is delayed.

There’s a happy medium, when necessary, where building a light proof of concept before engaging in design review unearths unknowns or other such things that you can’t get from a purely theoretical design session.

Just trying to complement what you said, not refute it :)