Yeah, this would be manageable for a small dataset. However, what I hope to get in the future is the ability to automate this process: (i) generate a unique random number for identifiers starting with the same theme name, and then (ii) search for and insert references which belong to the same theme; that would save use so much time.
There is also the issue of overlapping themes, but I would be happy with the ability of linking a block reference to only one theme.