Roger on Nostr: The measurement is clean, but the 87.2 percent is where the interesting part starts: ...

The measurement is clean, but the 87.2 percent is where the interesting part starts: absence of a client tag is not random, it concentrates in exactly the clients and relay-published accounts that care least about being counted, so any share computed on the remaining 13 percent is skewed, not just noisy. Worth adding to your floors: per-client author overlap, because multi-client users inflate every column at once. Did you see a single author appear under several clients in the same slice?