Wilbur

Wilbur will be the first data solution offered in our data marketplace and functions like an onchain firewall.

Wilbur is designed to address one of the most wasteful and pervasive issues in the industry: Sybil attacks.

A Sybil attack is a type of attack where a single adversary creates and controls multiple fake identities (or nodes) in a distributed network. In the context of crypto, Sybil attacks can impact consensus mechanisms, undermine token distribution (airdrops), and skew user analytics of onchain applications.

Fun fact: the Sybil attack gets its name from Sybil (1973), a book about a woman with dissociative identity disorder (DID). That woman's doctor, Cornelia Wilbur, was among the first to describe the disorder and inspired the name of this product.

Just like DID, Sybil attacks cannot be entirely cured, only treated. Our goal with Wilbur is to use state-of-the-art machine learning to better identify Sybil addresses onchain and at the very least increase the cost of spinning up Sybil addresses, thereby offering better protection to onchain applications and users.

Wilbur is based on our research around Reputation Oracles, a specifically trained feed-forward neural network (FFN) that classifies smart contracts as Reputable or Malicious with 94% accuracy, 87% precision and 85% recall. This work applied LongCoder to decompiled smart contract bytecode, which is a representation of the code of a smart contract. We are pursuing new approaches with Wilbur to generalize this classification not just of smart contracts, but also Externally Owned Accounts (EOAs).

This is where Pledges come in. Pledges enable users to connect and fetch data from the Web2 world. This connection of an address with a Web2 API is valuable, as it can prove humanity or non-Sybilness in a productive and private way.

Our inaugural Pledge application, Runner, allows users to set running distance goals and hold themselves accountable to these goals by sending money to an escrow that only returns the funds if, and only if, that goal is met. This entails a connection to a real-world sensor (e.g. Apple Watch) that would be cost-prohibitive for a Sybil attacker to replicate.

Users can still own the underlying data associated with their health goals by encoding it into a Vault, but it takes time and scale for this data to become valuable. Wilbur solves the cold start problem by making the connection between an address and a Web2 API valuable. That's because that association between an active blockchain address and a real-world API can then be used to train a model to classify any arbitrary address as non-Sybil.

By using the same approach as our Reputation Oracle, we hope to provide protocols with a way to programmatically deal with Sybils by using Wilbur. Proceeds from Wilbur are then distributed to the addresses that have used Pledges to contribute to its training data.