One thing that annoys me about the crypto space is the lack of tools for writers. There's t2.world on lens. And then there are several IP related protocols supposedly making it easier to copyright your creations and trade the copyrights.
Culture is bland because everyone is too connected. There is too much synchronization happening between people that articles either get no visibility or go viral. "What we need are pockets of cultural islanding" - protected spaces where ideas spread slower" (Seeing like a network)
One way to build these islands is to look for Cultural Vanguards, those who mint stuff before the mints become viral, those who discover thinking and ideas before the mass. We need to highlight the collections of people who are ahead of the curve.
While working on the OpenRank Mission at Bytexplorer my goal was to create a feed that highlights the casts and collections of Cultural Vanguards.
Full notebook on hex
Curating Cultural Vanguards
It's easy to explain conceptually what a Cultural Vanguard is, but harder to measure it. When you go from idea to metric, you loose nuances as you have to work with what you have. These are the assumptions I made:
A Cultural Vanguard supports creators with their money or their attention. People who have skin in the game show support to writers. The more money, the bigger your support. Not everyone has $ to support early creators and thinkers. Amplifying voices is another form of support. A proxy for amplifying voices is recasting something.
A Cultural Vanguard is not an influencer. Cultural Vanguards aren't influencers or trendsetters. Influencers are often too busy to pay attention to new and upcoming trends. The signal for these upcoming trends are weak. They are harder to catch. But Cultural Vanguards are often closely connected to influencer, via direct or indirect relationships. It is thanks to this close relationship that they can amplify new ideas and help them spread.
Data sources
To create an initial set of Cultural Vanguards I used three data sources:
Hypersub subscription: Hypersub lets creator create recurrent revenue through NFT subscription. This can be done for anything: Membership club, Apps, and also for writers. While this feed should focus on writers, I was lazy and just grabbed every hypersub that exists via a Dune query. For each subscription, I took the 10 earliest subscribers with the longest subscription. If you were the first subscriber, but only supported the artist for one month, you did not made the cut. In that way I accounted for "recognizing trends early" and "willingness to put $ down".
Paragraph mints: Of course, dealing with writers I created a Dune query and collected all Paragraph mints. No filtering here, as the dataset is very on-point for what I was trying to do, and relatively small. Everyone who minted is a Cultural Vanguard.
Farcaster discussions: Cultural Vanguards give attention and amplify writers on farcaster by talking about them and sharing their work. To keep it manageable I pulled all replies and recasts from a selected set of writer-related channels. Using this list the goal was to identify casters who have the "ear on the ground". I've tested three metrics and settled on Eigenvector Centrality based on it's speed of computation and my familiarity with the measure. Eigenvector centrality is a measure of influence that takes the global pattern of a network into account. It measures how far a node is to any other node. It's based on the first dimension of a factor analysis.
Calculating an OpenRank Score
OpenRank is a trust algorithm made by Karma. The purpose of the OpenRank score is to give each cast a score. Depending on the score a cast will appear higher or lower in the feed. As most people don't scroll to the bottom of a feed, this means casts with a higher score are seen by more people.
The algorithm takes two inputs:
Seed agents (pre-trust): Seed agents are users who you know behave according to whatever you are modeling. In my case I'm modeling for being ahead of the curve, so my seed agents are those who I know are Cultural Vanguards. Whatever your OpenRank score should represent, the seed agents act according to this. In my case, this list is made up of those early and committed hypersub subscribers, people who ming paragraph posts, and the top 50% casters based on their Eigenvector centrality.
Trusted agents (local trust): These are agents (aka case casters) who are in some way connected to seed agents. The assumption is that if I trust Anna to be a Cultural Vanguard, and Belle is in some way connected to Anna (e.g., Belle replies to Anna on Farcaster), then Belle is also a Cultural Vanguard. Some of Anna's skill to sniff out new ideas, rubs off on Belle. To create the set of trusted agents I combined the three graphs (hypersub minting, paragraph minting, and farcaster interactions).
Writers feed using the Cultural Vanguard ranking
Below is a snapshot of the writers feed with the original ordering of casts (3rd column from the left) and the new ordering of casts (2nd column from the right).
And this is how a feed with paragraph mints would look like using the Cultural Vanguard ranking
