The experimental design can be missing and that is generally not handled very well in the codebase. When I first did some experimenting for loading CELLxGENE datasets, I had to populate an empty design or else the UI would be broken.

I think we should make it non-null able and give it an initial empty value in the model declaration. Keep in mind that the column itself cannot be non-null because that table also stores other kind of investigations that lack designs.

Another thing to look into is adding a unique constraint on the EE foreign key to prevent a design to be used by multiple experiments. This can be achieved with a one-to-one relationship. It also applies to the MV relation and GEEQ scores.