The solution to Zooko's puzzle, as explained on the page, are "petnames." These petnames are basically the same thing that I am suggesting - there is never a truly ambiguous case that you cannot disambiguate. Even in the case when two separate records have exactly the same metadata you can disambiguate them by flipping a coin and adding an arbitrary a,b,c incrementer to one of them. In 99.9999999% of cases records won't be exactly identical, and some piece of the metadata can be incorporated into the key in order to disambiguate.

The crux of this thread seems to be: There will always be edge cases, therefore we must rely on meaningless unique hashes in order to index them. My solution is to create a key that gets almost 100% of all cases, and then create policies that disambiguate the rest by creating "petkeys" for them that are used instead of the default key system. This imbues almost all of our keys with meaning and uniqueness. It also makes it very clear - in the key itself - when there is ambiguity in a record.