I would say there's a spectrum. My guess is that whatever schema you could come up with could have an extra layer of metadata applied to describe the data to some benefit, but at some point you have to make a decision about how something is best represented.

To keep with the example of asking questions, if a sruvey is asked and each question is represented by field in a table for that survey, but they also want to track whether any particular question is answered, there are multiple ways to track this information. It can be done through a separate table (or multiple), it can be done with an extra boolean field per original question field, or it could be done by making each question field nullable, and ensuring that no question answered will ever be left null.

Which of those cases breaks the conceptual integrity of the domain modeling? I would argue that as long as they are applied consistently and accurately, none of them break it.