Artifacts for: Identifiability of Relational Queries in Multi-View Pretraining

The paper studies when a relational query is identifiable from a data-integration interface—that is, when every legal completion of the data consistent with the interface returns the same answer—and introduces two polynomial-time schema-design algorithms: CheckCert, which decides certification via an attribute-closure certificate, and Greedy-MinAug, which selects a near-minimum set of interface actions (resolvers, crosswalks, shared identifiers) that make a query identifiable.

The deposit has three parts: Experiment code (Python). Implementations of both algorithms, the full experimental pipeline (error-floor, capability-jump, minimum-augmentation, certificate-benchmark, scalability, and real-world witness-discovery experiments), and the builders for every dataset used in the paper—a synthetic benchmark with exhaustive ground truth, the BibInteg bibliographic dataset, and the real cross-source datasets CrossKG-DBLP (OpenAlex × DBLP), Amazon-Google, and Fodors-Zagat, alongside the WDC schema case study.