Credio: A Decentralized Risk Oracle Network for Real World Assets

1. Abstract

The market for tokenized Real World Assets (RWAs) is projected to reach between $3 trillion to $10 trillion by 2030. However, RWAs suffer from inherent oracle problems because on-chain asset issuance is being subject to credit and counterparty risks. These risks, probabilities of default of tokenised RWAs, present challenges that can increasingly be addressed through AI and advanced data science. We introducing Credio, a decentralized risk oracle network for RWAs, where machine learning inferences and predictions can be directly fed into smart contracts in an automated, decentralized, and privacy-preserving manner. The result is that tokenized RWAs are appropriately priced and monitored in many applications looking to scale on-chain RWA adoption by institutional investors.

2. Credio in DeFi RWA

Introduction to credit oracle

Untangled is developing Credio, a real-time credit oracle powering on-chain lending with AI-driven risk models and zero-knowledge proof. We build Credio on the conviction that much of RWA financing will be tokenized requiring robust credit infrastructure to scale. The credit oracle is a major missing piece in RWA DeFi today.

Health factor in DeFi lending

In any collateralised lending, it is important to establish the price of collaterals as it directly impacts the recoverability of the loan. This is the concept of collateral factor (value of collateral / loan amount). When collateral ratio drops to a certain level, say from 150% to 110%, collateral will be liquidated and the proceeds being used to repay the loan. 

All major DeFi lending protocols like Aave, Compound, Maker work in this way. However, in order to work out the collateral ratio, they need to know the current price of collateral. This is where price oracles such as Chainlink or Pyth come in. They provide price feed on a real-time basis so that the protocols can maintain their financial health through the monitoring of the collateral ratio 

The above protocols have one thing in common: their collaterals are crypto-native i.e. ETH or BTC. Whilst the collateral themselves are volatile in prices, there are deep markets for them, making prices highly observable. 

Current stage of RWA collateral monitoring

RWA lending protocols, like the above, also need to monitor their health through the collateral ratio. However, the difference is that RWAs, while less volatile, do not have on-chain or secondary markets where prices can be easily observed. 

So if these protocols want to value collaterals they need to use a ‘fair value’ approach based on credit modeling using statistical or machine learning approaches. These models generate ‘prices’ (or probability of default) for the credit collaterals which could then be fed into the collateral ratio above.

The issue is that not many RWA lending protocols do or have the ability to value and monitor these RWA collaterals. The lack of collateral monitoring and pricing is a major contributing factor to many defaults among many RWA pools to date. 

Credit modeling

In TradFi, the credit modeling or rating function is performed by rating agencies such as Moody’s, S&P and Fitch. Any public debt issuance will need to have a credit rating. In DeFi this is non-existent. 

A number of players have started to provide credit rating, modeling in DeFi. Untangled also built a machine learning based credit modeling capability that is providing credit prices for credit pools listing on Untangled protocol. This is a differentiation among existing RWA credit offerings in DeFi. 

Credit oracle

Just like Chainlink or Pyth is not the only source of crypto pricing, Untangled is decentralizing the credit modeling/pricing such that anyone with expertise in specific asset classes or the big rating agencies of the world can provide credit prices for RWA lending protocols. In order to feed those credit prices into smart contracts, an infrastructure layer needs to be developed. Where Chainlink or Pyth builds an infrastructure for feeding crypto prices, Untangled builds a credit oracle infrastructure to feed credit prices directly into smart contracts in real-time. 

Untangled also incorporates proof of reserve within Credio. With this bundled solution, RWA lending protocols not only know the price of the RWA collaterals but also their existence. 

The lack of on-chain monitoring a system validating the existence of the collateral and their prices was what stopped Maker and other leading DeFi protocols in lending to private credits RWAs but to focus on US treasuries instead. 

3. Solution

Credio addresses these issues by providing a decentralized oracle network that connects machine learning risk model predictions directly to smart contracts. This is achieved through an automated and privacy-preserving approach. The solution comprises three main components:

  • Shared Data Infrastructure: This combines off-chain and on-chain data, enabling modelers to build robust machine learning models and monitor ongoing performance efficiently.
  • Privacy of ML Models: Credio utilizes zero-knowledge proof (ZKP) technology to create proofs that accompany model inferences. This ensures that model outputs, verifiable on-chain, are derived from a trusted model using specified inputs. This transparency promotes participation from a diverse group of modelers, including rating agencies
  • Smart Contracts Coordination: A set of smart contracts coordinate the model inference feed with consumer applications (dApps). 

4. Key Actors in Credio Network

Consumers: These include smart contracts/dApps, individual or entity consumers of machine learning predictions, such as risk parameters, credit ratings, or risk scoring. Consumers set challenges to build relevant ML models, provide performance targets, call to update risk parameters, and pay network fees.

Modelers: These participants respond to ML challenges to build predictive models such as credit ratings or scoring that meet performance standards. They also provide ZKP along with model inferences/outputs.

Validators: Validators play a crucial role in the network's security, ensuring models meet performance benchmarks, verifying proofs, and validating off-chain data inputs into models.

5. The Credio Workflows

Model Building

  • Consumers: Define risk parameter inference needs, performance criteria, and reward for model building.
  • (1) Modeling: Modelers enter challenges, build models leveraging data infrastructure services, and convert models into zkML models using the ezkl library.
  • Validation: Validators ensure the models are high quality, meet performance benchmarks, and that outputs are genuine given a set of inputs, without revealing model mechanics and proprietary information.

Model Consumption

(2) Pull Architecture: Calls to update risk parameters can be triggered based on time or events.

(3) Data Inputs: Both on-chain (e.g., token ID) and off-chain metadata are submitted as inputs to a trusted model.

(4) Inference and Proof Generation: The model generates inferences based on the inputs along with ZKP. Consumer submits inferences and proofs to oracle coordinator contracts via the Pool.

(5) Off-chain Validation: Validators validate model performance, watch call data and verify off-chain data and ZKP.

(6) On-chain Verification: Oracle coordinator contract verifies proofs and updates smart contracts with new risk parameters.

6. Tech Stack

Oracle Contracts

  • EVM-based smart contracts deployed to Celo, a layer 2 blockchain. Read more about the oracle contracts and how to integrate it, refer to Appendix 1.
  • Integration with PoR. Read more about how we implement a proof of reserve with Merkle Tree, refer to Appendix 2. 

Zero-Knowledge Proof

  • Conversion of ML models to zkML models using the ezkl library.
  • Proof generation (single or aggregated).

For more information on zkML work flow and technology please refer to Appendix 3.

Data Analytics Platform

  • Data Warehouse: Real-time data download, decoded, and transformed into easy-to-use business tables.
  • Query Engine: Supports efficient data queries.
  • Analytics Dashboard/BI Tool: Provides user-friendly business intelligence.

Account Abstraction

  • Account Abstraction: ERC 4337 enables users to sign up and create non-custodial wallets with just an email address.
  • Paymaster: Covers gas fees on behalf of users.

7. Use Cases

  • Private credits: credit oracle and collateral monitoring for tokenized private credits and on-chain securitization.
  • Stablecoin: Connecting rating agencies’ rating outputs with protocols that issue or use stablecoins.
  • Counterparty Risks: Credit value adjustments (CVAs) for any transactions involving real-world entities, such as issuers of tokenized RWAs or RWA-backed stablecoins.
  • DAO bond issuance: rating for bond issuance by DAOs.
  • Digital bond issuance: rating for digital bond issuance by TradFi

8. Business Model

Inference Feed

  • Oracle Fees: Similar to Chainlink but focusing on high-value, low-volume feeds.
  • Fee Distribution: 60% to Modelers, 20% to Validators, 20% to protocol treasury.

Data Analytics Products

  • Free Service: Obfuscated datasets and certain dashboard usages.
  • Premium Services: Training datasets and analytics dashboards are charged.

9. Traction

Untangled Finance, the developer of Credio, is backed by Fasanara, an institutional asset manager with $4 billion in assets under management (AUM). Untangled Finance also develops an RWA lending protocol focusing on institutional-grade assets. This partnership helps bootstrap collateral pools with 140 asset originators in over 60 countries. The first pool is live on the Celo mainnet, with Credio providing risk oracle, credit analytics and monitoring services. 

As these RWAs are tokenized, Credio provides risk rating and monitoring services. A proof of concept (POC) is being implemented with a top rating agency as the modeler, Fasanara’s diversified pool as the consumer, and Credio as the risk oracle.

10. Roadmap

Data platform

  • Data sourcing: connect with more off-chain data providers   
  • Data analytics tools for credit risk assessment and monitoring

Credit oracle

  • Credio CLI for model interactions and coordination during challenges
  • Smart contracts for running machine learning challenges and monitoring risk parameters  

Go-to-market

  • On-chain securitisation of private credits and other asset backed financing, ie. on-chain asset backed securities
  • Digital bond issuance by TradFi or DAOs 

11. Conclusion

Unlike native crypto collaterals, RWA collaterals carry significant credit risks. The development of RWA DeFi has been hampered by issues such as bad debts and non-performing counterparties. Currently, there are no oracle solutions for illiquid RWAs. Even liquid collaterals, such as tokenized money market funds, are plagued by counterparty risk, leading to differences in returns whilst being based on the same underlying collaterals. Without a robust system for risk pricing and monitoring of tokenized RWAs, large-scale institutional participation remains challenging.

Credio is an oracle network that helps DeFi protocols, issuers and investors to manage credit risk regarding RWAs in an automated, decentralized and privacy-preserving manner. We pioneer in the fast growing RWA segment within the oracle market. We are similar to Chainlink and Pyth but we focus on providing machine learning inference feeds for illiquid RWAs from a decentralized community of data scientists, risk managers and rating agencies. By blending AI with blockchain, Credio brings a new predictive layer to smart contracts, enabling the massive growth potential of the tokenized RWA economy, by making creditworthiness programmable.