Often, imo, this depends on where you want to put the validation complexity. You can move validation to the front end and "prevent constructing bad states" through a combination of good UI and front end data validation, i.e. validate all the pieces of an entity to make sure the eventual Entity will be valid. But the opposite trade off would be to keep your front end simple, and have the data represent anything that is possible in the UI, and then you can validate anywhere in your stack that you want or where it makes sense for any engineering reason. You can still validate the UI models on the front end before sending them over the wire. You can validate at the endpoint and return HTTP codes. You can run logic in your database to massage/correct/fix the data any way you want. You could even throw things into a queue for post-processing later on, or send to another server for processing, whatever you want.

In theory it sounds great to say "just prevent construction of bad states," but this is a complex problem in professional software that involves coordination between programmers, designers, and product managers to get the UI features into the right place where preventing the construction of bad states doesn't lead to a frustrating UI experience for users. It's often far simpler, more pragmatic, and less complex in terms of coupling in both software and humans to just accept whatever state the UI gives you and figure things out somewhere else where you aren't bothering your users so much.