Signal Qualified Cross-References as a Verification Method: The Normative Environment of the EU AI Act

Summary

This paper presents a provision-level model and construction protocol for 'qualified cross-references' in legal knowledge bases, developed through a bilingual corpus of fourteen legal instruments surrounding the EU AI Act (Regulation 2024/1689). It argues that a mere link between legal provisions is insufficient; a curated system must also state the legal character of the interaction, identify supporting provisions, preserve conditions, and remain consistent when read from either direction. The methodological contribution is 'bidirectional inversion,' where a relationship documented from one act is reconstructed from the other's perspective as a verification step. Applied to the AI Act corpus, the protocol surfaced six incorrect article references, three inaccurate legal qualifications, and one divergence between two published descriptions of the same interaction, including a case involving Regulation 2019/881's cybersecurity presumption. The paper offers both a map of one regulatory environment and a reproducible method for building inspectable, internally testable legal cross-reference knowledge bases.