Outbox Bootstrap relays (i.e. purple pages) could be a good way to get an idea of these follows lists, but yes, agreed it’s an unsolvable problem. (Well, problem is an overstatement; it’s a stupid legacy metric that was always bullshit in centralized platforms)

nostr:npub180cvv07tjdrrgpa0j7j7tmnyl2yr6yr7l8j4s3evf6u64th6gkwsyjh6w6  had a cool idea about using probabilistic COUNTs based on HyperLogLog to aggregate large sets of data across multiple relays without downloading all the data multiple times and having an estimation of the duplicates.