An Explainable Methodology for Company Inspection Prioritization: Evidence Integration, Provenance-Aware Graphs, and Policy Sensitivity

Integrating heterogeneous administrative records into explainable inspection priorities requires explicit handling of missing information, uncertain entity links, and source provenance. This study develops a computational methodology that combines source-specific evidence processing, a provenance-aware evidence graph, and transparent rules for company selection and prioritization. The methodology is implemented and retrospectively evaluated using Slovak administrative records covering three years, provided by the Ministry of Interior through collaboration with the Institute of Administrative and Security Analyses (ISBA: The Inštitút správnych a bezpečnostných analýz is an advisory and analytical unit of the Ministry of the Interior of the Slovak Republic). Registry, financial, debt, value-added tax (VAT), address, and company-role records are processed separately under explicit data-quality and missingness rules before integration into company-level selection and ordering decisions. Transparent rules identify review candidates, while historical liquidity and asset information guide operational ordering. A provenance-aware evidence graph links observations and assessments to their sources while preserving uncertainty in identity matching, temporal coverage, and missing information. Developed and evaluated in Slovakia, an EU Member State, the methodology offers a basis for adaptation elsewhere, particularly in other EU Member States, subject to adjustments for local data sources and institutional requirements and validation in each setting. Applied to 482,902 registry companies, the methodology produces company-specific review explanations. Exhaustive reconstruction checks reproduce the adopted patterns, selection decisions, and ordering without discrepancies. Joint restrictions on relationship and VAT evidence retain 93% of baseline top-500 membership. Written feedback from ISBA reports documented corporate impersonation in 4,270 of 5,000 expert-selected companies reviewed from the first priority group, a confirmation proportion of 85.4%; the remaining 730 cases are unresolved. This non-random, unblinded review does not establish population-wide accuracy or comparative effectiveness. The study contributes an auditable methodology supported by computational reconstruction, conditional policy robustness, and institution-reported case-review findings.