Even disregarding all the deficiencies with BIP110 which have been thoroughly elaborated upon elsewhere, I have no faith whatsoever in those numbers being representative of anything but a handful of people spinning up nodes to pad the numbers.
First of all, of the 1905 nodes registered on Bitnodes that advertise BIP110, more than 1300 are onion nodes. Creating an unlimited amount of onion addresses and pointing them at a node, then submitting that address to Bitnodes, is effectively free. As such, if anyone were to accept use Bitnodes stats for arguments of node distribution - and that’s a big if - the only meaningful number would be counting the nodes that have some semblance of cost, which would limit it to IPv4 nodes. Which would bring the number down to ~420 out of 7242, or 5.7%
Secondly, even those numbers do not at all match the distribution of subver strings that connect to my four public listening nodes. As of right now, I’m seeing:
- 1026 nodes running Core
- 141 nodes running Knots without BIP110
- 16 nodes running Knots with BIP110
So 16 out of 1167 nodes are advertising BIP110 in the subver, or 1.4%. Assuming that all nodes have an equal number of outgoing connections and an equal chance to connect to any of my nodes, this statistically gives a greater than 99% confidence that the real number is between 0.5% and 2.3%.
I put a raw data dump on pastebin for anyone interested. I filtered out all subvers without “Satoshi” in the string, which mostly limits it to Core and Knots nodes.
There is a reason soft forks are activated by mining power alone; node listen counts will always be suspect, and unlike node sockpuppeting, mining power cannot be falsified.