zScore: A Universal Decentralised Reputation System for the Blockchain Economy

TL;DR

zScore introduces a universal, decentralized reputation primitive for the onchain economy by quantifying wallet credibility from active onchain behavior. The framework fuses onchain data with state-of-the-art neural networks in a multitask setting, guided by clustering to define zScore intervals and cryptoeconomic security via EigenLayer AVS and Merkle proofs. A casestudy on Aave V3 demonstrates that higher zScores correlate with healthier borrowing and repayment, while outperforming prior proprietary approaches in robustness and transparency. The work envisions broad impact across lending, DEXs, and other DeFi verticals by enabling dynamic LTVs, personalized rates, and reputation-driven incentives that improve capital efficiency and security in a trustless environment.

Abstract

Modern society functions on trust. The onchain economy, however, is built on the founding principles of trustless peer-to-peer interactions in an adversarial environment without a centralised body of trust and needs a verifiable system to quantify credibility to minimise bad economic activity. We provide a robust framework titled zScore, a core primitive for reputation derived from a wallet's onchain behaviour using state-of-the-art AI neural network models combined with real-world credentials ported onchain through zkTLS. The initial results tested on retroactive data from lending protocols establish a strong correlation between a good zScore and healthy borrowing and repayment behaviour, making it a robust and decentralised alibi for creditworthiness; we highlight significant improvements from previous attempts by protocols like Cred showcasing its robustness. We also present a list of possible applications of our system in Section 5, thereby establishing its utility in rewarding actual value creation while filtering noise and suspicious activity and flagging malicious behaviour by bad actors.

Figures (7)

• Figure 1: Overview of the mechanism of assigning users zScores The graph represents users’ onchain history embedded in a 2-dimensional space in which we have been able to separate users into three clusters; each of these clusters has zScore bounds [Section 3.2], and the neural network uses this cluster data along with the user features to predict the user’s zScore [Section 3.3,2]
• Figure 2: A flowchart of the model architecture We first transform the user's onchain history into relevant features described in [Section 3.1]; we then classify the user into a cluster [Section 3.2]. The features are then scaled and fed into the neural network [Section 3.3]. The output from the scoring head is then used to scale the score according to the bounds.
• Figure 3: Flowchart of zScore Execution service zScore AVS, we first fetch user logs and then extract features from it before passing them to the model [Section 3.3]. We then store user features and their zScores in a DB, which we then generate the Merkle root for. The Merkle root is then validated by the validators, and once quorum is reached, we publish it to Base [Refernce].
• Figure 4: Distribution of users across clusters Left - the distribution of users with non-zero liquidations. Right - the distribution of users with zero liquidations. We have assigned the subclusters of cluster0 IDs from 0-18 [Section 4.2]
• Figure 5: Cluster-wise distribution of users with non-zero liquidations [Section 4.4] Almost all clusters have skewed distributions, with clusters having few users converging to a small range of zScores. An important insight, however, is the convergence of zScores of new users to a range between 100-250.