Sorry, what I meant was there are many cases where throwaway keys are used which are not scoped within NIP-17/Marmot/etc and would still fall through into your w0 cohort I also suspect there’s lots of “probing” going on- people making bots and testing if they get past certain relay/WoT filters for example. I think a good chunk of this can be filtered out by simply ignoring any key that doesn’t last more than some minimum amount of time. I proposed a month. No idea what the “right” number is. But this would get a more accurate reflection of retention than the current methodology.