I was about to say the same thing - if your set of conditions are changing, it sounds like you want a dataset to populate the values, and can just do another join to get the filter behavior. You can even have this be a writeback dataset so you can build out a nice workshop flow for opening / closing a plant.
Also yes, if you have an embeddings column and you’re trying to do nearest neighbors between the data represented by the embeddings, it’s almost always going to be best to do the distance measure on the embeddings.