The pagerank of crypto is here, allowing any developer to provide contextual and personalized feeds that capture the ever-expanding universe of onchain and offchain data.

It's called OpenRank, and I've created an end-to-end example flow using their python sdk and Dune queries.

Base

In this dune dashboard, you can see the recommended mints, tokens, contracts, and farcaster channels weighted by the outputs of any openrank graph that's been created. 

In my example graph, I used all /base channel casters and their token transfers.

headmaster @duneanalytics, previously mirror and consensys

In this dune dashboard, you can see the recommended mints, tokens, contracts, and farcaster channels weighted by the outputs of any openrank graph that's been created. 

In my example graph, I used all /base channel casters and their token transfers.

https://dune.com/ilemi/openrank

There are two key concepts to creating your OpenRank graph. Pretrust and localtrust. Think of localtrust as your measurement of interactions between two nodes, and pretrust is your "seed" selection of nodes who you think should be more trusted in the network.

The /base channel has 60,000 casters who have transferred tokens at least once. I aggregated their stablecoin and ETH transfers on optimism, ethereum, and base to get my "localtrust" graph. To get my pretrust nodes, I sorted for the top casters by engagement in the channel.

From there, we want to run a quick check on coverage/connection of the network. This can vary greatly depending on the sybil nature of the graph, but in our cases 19% of the localtrust graph nodes can be reached from 100 seed nodes!

Next, the openrank algorithim is run. It's a fairly simple formula, based on eigentrust. "P" is your pretrust and "S" is your localtrust, and "c" is a weighting of the trust value between the two.

These localtrust interaction graphs are typically exponentially distributed, so having good pretrust nodes actually helps normalize the distribution of trust.

That sounds complex to build and run, but it's really just 10 lines of code. It's as easy to use as scikit-learn in the ML world.

After we have the results, we can quickly connect it to the Dune API with the "trending contracts" feed to see the top contracts being interacted by casters in the /base channel. I've applied a simple weighting of "v*number of calls" to get a contract score ranking.

bytexplorers

I've only shown a simple graph here, there will be a /bytexplorers mission soon for those who want to be more hands on! 👀

I'm excited to help bring @openrank to life, please check out the notebook and play with graphs yourself.

I've only shown a simple graph here, there will be a /bytexplorers mission soon for those who want to be more hands on! 👀

I'm excited to help bring @openrank to life, please check out the notebook and play with graphs yourself.

https://app.hex.tech/e8198fd2-0779-4dff-a368-06371d3af467/hex/fa721eba-e774-4036-bda1-52323202ef9a/draft/logic?view=notebook

OpenRank is like Dune but instead building a community around valuable network graphs. They manage the compute and devops, so anyone can come in with input graphs and train/store without setting up any infra.

Let's walk through my example graph creation process next!