Coverage and Structural Relations of Three Agentic AI Risk Taxonomies Across 131 Documented Incidents — Supplementary Dataset, Codebook, and Analysis Scripts

Supplementary materials for the manuscript "Coverage and Structural Relations ofThree Agentic AI Risk Taxonomies Across 131 Documented Incidents" (submitted toDiscover Artificial Intelligence). Version 3, accompanying the third revision. The deposit contains 131 agent-related AI incidents drawn from the AI IncidentDatabase (2014-2026), each coded against the OWASP Top 10 for Agentic Applications2026, the Microsoft AI Red Team taxonomy of agentic failure modes, and the NIST AIRisk Management Framework. Files: the coded workbook (XLSX and CSV); a codebookdocumenting category boundary decisions; the reconciliation log; an independentblinded coding pass over all 131 incidents in two conditions, with both coders'per-incident output and rationales; sensitivity analyses for the pre-2022 subset,the keyword threshold, the failure-only subsample and the strict single-limbinclusion criterion; a blinded coding kit allowing an independent human coder toreplicate the reliability check on the prespecified 25 percent subsample, with ascript that scores a returned sheet; a redacted copy of the May 2026 scopingdocument with the deviations from it tabulated; fourteen Python scripts reproducingevery reported statistic and all seven figures; and the figures at publicationresolution. Incident source URLs and access dates are included for every record.Incident data derives from the AI Incident Database (incidentdatabase.ai),licensed CC BY 4.0.