Erasure Thesis

Note: Placeholder holds Erasure’s NMR.

Information is a strange good: it’s created in unpredictable ways, you can’t know if it’s good until you have it, and it can be reproduced at no cost by anyone who does have it. This makes it hard for markets to price and distribute its value. But we can use crypto to address these challenges. Erasure is a new protocol for exchanging valuable information on the internet. It uses encryption, smart contracts, and staking with the Numeraire token (NMR) to create a trusted, decentralized venue for exchanging data. 

The protocol is built and used by Numerai, a hedge fund managed by an anonymous network of data scientists. Numerai’s community competes to build machine learning algorithms on the fund’s data to generate stock market predictions. To earn money by submitting predictions, users have to “stake” some NMR by locking it in a smart contract. Numerai burns the stakes of bad predictions and rewards good ones with more NMR proportional to the size of their stake. So the more you risk behind your data – i.e. the more skin in the game you have – the higher your potential reward. 

Based on current market prices for NMR, Numerai has paid over $20 million to its data scientists.

Based on current market prices for NMR, Numerai has paid over $20 million to its data scientists.

Numerai’s breakthrough is proving that staked data is better than un-staked data. Users who stake more behind their submissions are more likely to be right than users who stake less. This helps the fund decide how to allocate its capital based on anonymous predictions. Numerai has paid millions of dollars to thousands of users with this system, proving people all over the world are not only willing to produce and sell valuable information in exchange for crypto, but also to put real value at risk behind their claims.

Staking proved to be a brilliant quality filter for information. But Numerai was a centralized service focused only on financial data. And if this idea worked for the stock market, it could probably work for any kind of information. So they decided to turn this mechanism into an open protocol, and that’s how Erasure was born. 

The Protocol

Erasure consists of three primitives: payment, recourse, and track record. The first primitive facilitates the exchange of information. A user looking to buy some information can stake behind their request to prove they are serious about their request, and anyone can sell them the data provided they also stake some NMR behind it (according to the buyer’s stake requirements). The amount staked tells consumers how confident the producer is in the quality of their information, and the data can be encrypted such that only its buyers can read it. Everything is automatically recorded on the Ethereum blockchain, so the history of the transaction cannot be removed. 

Crypto makes payments easy; because the exchange is managed by smart contracts, each party can trust the information will be paid for and delivered as posted. It’s also possible to stake and post data to Erasure without encryption and for zero payment, solely for the purpose of staking a permanent claim on the blockchain. 

Recourse is what allows the parties in a transaction to threaten the other’s stake after the exchange. The first kind of recourse available on Erasure is called punishment. Users can set a “punish ratio” for their stake. For example, if a producer sets it at 1:9, it means the consumer can destroy 9 NMR of the producer’s stake for every 1 NMR they destroy of their own. A consumer may grief a producer for bad information, like a wrong price, fake news, spam, etc. But it costs money to burn money, so people are discouraged from abusing the feature. The mere threat of grieving is enough to keep most people honest (though it’s possible to be both honest and incorrect), but in general, a high griefing ratio means the producer is very confident because more of their stake is at risk. 

Finally, there’s the track record, which is composed of every user’s history. It’s Erasure’s equivalent of eBay or Amazon reviews, except you can trust it because it’s made entirely out of blockchain data with economic guarantees. The meta-data generated by an Erasure transaction provides a lot of information for producers and consumers to decide if they can trust each other. Anyone can see the record of transactions for every user, including whether they’ve been burned or not. In particular, a user’s “punishment history” provides a useful measure of quality without revealing the contents of the information being exchanged. A user with a large amount of NMR staked and a long history of not being punished is a lot more appealing than a user with the opposite qualities, and so on. 

With these elements, any Web3 developer can make a stake-based information market for their application. Each primitive is extensible and composable, meaning app developers can customize anything from data formats to payment methods, invent new recourse mechanics, and design their own reputation formulas based on track record. For example, you could extend the payment primitive such that staking happens with NMR but payment with another token (like ETH, Dai, or your own), or you could create a recourse template where stakes are burned based on what happens in other smart contracts, such as the outcome of an Aragon DAO’s vote or the output of an oracle. 

Erasure Today

Numerai moved itself to Erasure when it launched on Mainnet, and released a new app Erasure Quant to leverage some of the new functionality. All of Numerai’s activity now runs on the new protocol, and we can see activity growing by looking at how much NMR is staked in Erasure contracts over time. According to Defi Pulse, there’s ~70,000 NMR “at risk” staked behind predictions, worth about $1.2 million at current market prices (~$16/NMR):

https://defipulse.com/erasure

https://defipulse.com/erasure

Developers can use the Erasure SDK wherever staking can help increase the quality of information. You could, for example, host a Numerai-like app that uses Erasure and 0x to run an entire crypto trading operation. Or you could build a calendar where accepting an invitation requires each party to stake behind their commitment to attend. Or an email client where you can burn spammers. Or a news feed that ranks stories by stake. Even ads could be staked.

But it’s also important for end-users to use Erasure directly and understand how it works in order to trust it. This is where Erasure Bay comes in: 

ErasureBay is the first consumer application built on Erasure.

ErasureBay is the first consumer application built on Erasure.

Erasure Bay is a consumer interface for buying and selling any kind of information. Anyone can post a request for data, with custom terms such as the amount they’re willing to pay, how much the seller has to stake in order to submit an answer for consideration, and what the “punishment” rate and period are (i.e. griefing). Erasure Bay uses Twitter for identity and the dollar-stablecoin Dai for payment, but burns NMR in the background to make it simple for regular crypto users. Erasure Bay showcases the utility of the protocol and the kinds of apps you can build with it. 

Data Legos Against Bad Information

Given its heritage, it’s tempting to think of Erasure as a financial protocol. But it’s bigger than that. Michael Senna, co-founder of 3box, introduced me to the term data legos. Like DeFi’s money legos, Web3 data protocols serve as components for thin applications which are composed out of multiple crypto-services. Erasure sits at the intersection of money and data legos. Money is data, after all. 

Erasure solves the problem of bad information online. What would Reddit look like if users had something real to lose for every post, would troll farms be able to manipulate it as easily? What if you could burn a publication’s stake for promoting fake news, or an advertiser for misleading consumers? Can Erasure help manage the threat of deepfakes

As the sheer amount of information online becomes unmanageable, we need new tools for filtering and assessing its quality. A system for validating data based on the history and stake of its producer is a fascinating proposal. It may be that the only way to distinguish good from bad information online comes down to how much value its creator stakes and their track record. That’s the vision of this protocol. And once you understand how it works, and its potential, it’s easy to see how it fits everywhere.