Could you make your content-based spam reduction ML open source?

We’ve got good diversity in relays at the moment. If we separate relay security and make that global we can retain the optionality for implementations to keep the network strong.

Something like how pihole lists work, relays could access security modules from different people doing different things.