Paper 2026/1315

VERDICT: A Cryptographically Verifiable Framework for Secure Data Lineage in Decentralized DAGs

Maria Potop Butucaru, Sorbonne University

Abstract

Ensuring the integrity and traceability of data transformations in distributed systems presents significant challenges, particularly in environments where data privacy and decentralization are paramount. This paper introduces a novel secure lineage verification system based on Directed Acyclic Graphs (DAGs) and homomorphic hash functions, VERDICT. Our approach represents data artifacts and their transformations as two interconnected DAGs: a Data DAG tracking data dependencies and an Event DAG capturing causality between transformation events. We propose a level-based DAG compression technique that decomposes these graphs based on distance from genesis nodes, enabling efficient verification through skip DAG structures. The system incorporates bucket indexing and Merkle tree verification to provide cryptographic guarantees of data and event existence. We present formal algorithms for DAG construction, level-based hashing, skip DAG traversal, and verification processes. Security analysis demonstrates the system’s resistance to tampering and modification attacks while maintaining privacy. Replay attacks are prevented through an application-layer challenge-response mechanism. Our approach has significant applications in federated learning environments and decentralized architectures, where it can serve as a notary component for tracing events without compromising data confidentiality. Theoretical analysis shows that our method achieves verification in 𝑂(log𝑛) time in the number of DAG levels, independent of the number of nodes per level, making it suitable for large-scale distributed systems.