ReplyGuy Is evolving, a positive stimulus for relay and clients to improve their filter capabilities.
Discussion
Para quĂŞ servem?
Yes but it’s still all whack-a-mole. The best filtering is on the network level and for aggregators that don’t have access to the connection info it is challenging.
The only long term solution I can come up with is npub reputation where you face very strict rate limits initially.
I think ratelimit would not help for disposable npubs - I mean ones that are generated for single event. If network is flooded with unique npubs ratelimiting is not really useful
Well it depends. If you’re running the public relay that is being spammed, you can see what IP(s), ASNs, user agents, etc the spammer is using and rate limit on the network side.
If you’re an aggregator like filter.nostr.wine that doesn’t have that network view, then it’s MUCH more challenging. You need to focus more on the content/pattern of the messages and reputation of pubkey posting them.
I agree that we should start restrictive, a first filter by WoT should be sufficient, then we can mix other variables.
Also the UI should adapt, for example temporary hiding instead of filtering out completely, to let the user dig in false positive and unlock them.
Yup, I’m sure we will get there. UI is very important here but also thinking about the relay side.
We need to prevent events/purge them, not just keep them from being seen otherwise spam becomes expensive to host eventually.
Yes, I talked about the client perspective but the relay part is also really important, of course. As you say, at relay side things are harder since they don't have direct access to the social graph.
The usual proposals here seem to point to PoW, a payment or an invitation, to unlock access. But I fear that all these approaches could create a lot of friction and slow down onboarding.
Even network filtering does not seem very effective, when you have disposable key pairs and VPNs.
I would avoid to delegate to AI this job, this would really escalate to an AI vs AI fight.
Do you have any other ideas?
I think it’s going to be a challenge. I am actually fairly bullish on the network filtering side as part of the solution but unfortunately can’t use it for our aggregator. There are ways to circumvent it but it is easier to fingerprint connections than you might think, even with multiple IPs/ASNs.
I think relays will use social graphs based tools a lot but I’m not sure how to deal with new users. They will end up being very hard to discover if we apply strict rate limits/reduce their visibility until they gain followers/engagement.
We could encourage them to pay to “skip the early reputation line” but I don’t think PoW will prevent spammers. There needs to be a way to onboard without cost though. Perhaps some type of vouching from existing “reputable” users. Still can be gamed though. Lots to consider.
This seems a good starting point:
Invites, paid, to relays and suggested follow lists? This is more on the client though.
Thank you for caring for it.
It made my relay less lonely.