Before this conversation forked off in a direction I didn't want it to go, I'd like to thank everyone, privately and publicly, that gave me a hint as to the distribution of /25s and greater in their networks.

I was at the time, trying to get "libreqos.io" to crack the 32k customer barrier, which we did soon afterwards, and to get a decent histogram for our trie emulations of what real customer traffic might look like once we started scaling past 40Gbit. (the XDP trie lookup is pretty expensive for us currently, but a lot cheaper than the /32s)

Some idea of failover mechanisms other than BGP (igps are all over the map) came from this conversation also. I would certainly like to learn more about how well that works.

A new question I have... is there any decent rules for CGNAT vs bandwidth vs customer allocations of ipv4 space? I've had a few horror stories shared with me privately about that, of note was 9000 *fiber* users behind a single /32. with no way to haul ipv6 or more ips there....