Njal on Farcaster

Content pfp

https://warpcast.com/~/channel/farcasterunion

0 reply

0 recast

0 reaction

Njal pfp

'High signal' love how they use this word. Everyone want to be a high signal user, and everyone want to read high signal content. I have no Idea what it means. Casters with Pro seems to be high signal, and who knows who else. If your high on the rewards board it will be considered high signal content. I can speculate, but it's still untouchable, a container concept. Love it. https://farcaster.xyz/dwr.eth/0x9a3bca77 https://farcaster.xyz/deodad/0x3fb1f8e9

2 replies

1 recast

3 reactions

Pichi pfp

It’s worrisome. Yes, many of the folks who subscribed are super supporters of Farcaster. But many are speculating on the NFT. So it’s a signal, but not THE signal.

1 reply

0 recast

1 reaction

Njal pfp

It is concealing. If pro subscribers are high signal users (regardless of the spam label) then it means that it is only about turnover, there isn't any other use for them than Pro subscription payments. Mini app use by spam label 0 does not hinder the rest of Farcaster (as far as I know), so there is no reason for FC to hinder them in that. In my opinion, high signal will be anything that (in)directly maximizes profits; being on the network longer, attracting new users, making more transactions (resulting in more swabs), etc.

1 reply

0 recast

1 reaction

Pichi pfp

I love that so many devs are looking at it closely and trying to add more data layers. It feels like we finally have enough legos. Neynar scores now include mini app usage to catch folks who are interacting differently. Spam algo is binary now. Pro badge shows who has money. OpenRank shows network connections @geoffgolberg is highlighting inauthentic accounts It feels like we should be able to Captain Planet this shit. I know @compez.eth has been working on this for months too. We don’t want to leave people out, but don’t want to reward our farmers and extractors. Feels so close.

1 reply

0 recast

3 reactions

C O M P Ξ Z pfp

I think there are many complex criteria that we can use for evaluation. But people don't like to get involved with multiple conditions! This is usually very overwhelming for accounts that are newbie! So we need to consider a general criterion. I'm not sure yet but I think I can introduce a general evaluation index in right time that don't just rely on FC! Tbh, playing with data is really time consuming! I would love for @rjs to help us with this.

3 replies

0 recast

3 reactions

Njal pfp

I researched @mvr's and my dataset about the 1st 10k pro subscribers in a slightly different way than before. Both FC and Neynar measures spammy behaviour, OpenRank doesn't. So, FC and Neynar could have consensus or disagreement about who's a spammer (in this case I simplified the Neynar score to: <0.3 = 0; >0.7 = 2; 0.3–0.7 = "no label") In the case of consensus, the outcome is clear. In other cases, OpenRank could help to give casters with a better OpenRank score the benefit of the doubt. Of course, it should be stricter on a no label | 0 combi than a no label | 2 combi. However, this could end up with more "no label" values than FC. In this case FC has 2,307 no label values (most of them new accounts btw); Neynar score has 6,416 times a score between 0.3–0.7. This is an arbitrary bandwidth btw, however making it smaller isn't satisfying (a 0.45–0.55 bandwidth change the consensus with a factor of 1.25). Imo consensus doesn't need further judgement (except from a trusted user), but the rest does.

1 reply

1 recast

3 reactions

Pichi pfp

Neynar recently did an update to include minimal usage. This gave a bunch of folks a bump. But accounts that post no original content and just mini apps to their timeline probably get marked as spam by Farcaster. I wonder if that explains some of the divergence. This really is shaping up to be a strange cohort of users!

1 reply

0 recast

2 reactions

Njal pfp

When there isn't consensus about seeing a spammer, it's 40 times that FC says that they are a spammer, and 674 times that Neynar says that they are a spammer. Between no label, and not being a spammer it's 3 times that FC says no label, and 3,500 times that Neynar has a score between 0.3–0.7. Between no label or being a spammer, it's 401 times that FC says no label, and 1,013 times Neynar that has a score between 0.3–0.7. So, it doesn't look like that FC is stricter on mini-apps-casters. Changing the bandwidth to 0.45–0.55 for the Neynar score doesn't make it much better because: 40/674 changes to 189/2,024 (0 vs. 2) 3/3,500 changes to 55/1,363 (no label vs. 2) 401/1,013 changes to 511/412 (no label vs. 0)

1 reply

0 recast

2 reactions

Njal pfp

Based on a new follower with a Twitter account with 6 numbers at the end (always spam imo), it could be that this difference is a loophole used by spammers to get spam label 2. Connect Twitter account, and wallet with $25+ (if $120,- isn't a problem, this isn't a problem too) + mini app posts + some handmade reactions. Will check later today (probably morning in the US) some accounts in the dataset if this is the case too.

1 reply

0 recast

1 reaction