Supplementary evidence record for The AI Incident Actionability Gap: Public Evidence and Operational Interpretability in AI Governance

Version 1.2 follows a targeted consistency and source-integrity audit of the final analytical record. The audit corrected 24 case-field codings across 14 retained cases: 23 cells were changed from Partial to Present where Version 1.1 had applied evidential requirements not contained in the field-specific decision rules, and System identification for the UK judicial-warnings case was changed from Present to Partial because the public record established actual or suspected generative-AI use but did not identify a specific tool consistently across the joined cases. One case was removed because the retained public evidence documented an AI-tool restriction but did not establish AI involvement in the underlying incident at the threshold required by the study’s operational definition. The final analytical sample therefore contains 18 cases. Associated coding rationales, ambiguity notes, evidence-environment metadata and aggregate counts were aligned. The Replika evidence record was corrected to replace an erroneously linked Garante decision concerning ChatGPT with the corresponding Replika decision. The Porcha Woodruff Associated Press title, date and URL were reconciled, the Google Gemini source title was aligned with the retained page, the Solicitors Regulation Authority source metadata for the UK judicial-warnings case were aligned with the official 2023 Risk Outlook report, and the Public Law Project evidence metadata used in the DWP cases were aligned with the Public Accounts Committee’s 2025 Tackling fraud and error in benefit expenditure 2024–25 inquiry. The coding guidance for Oversight-route clarity was also clarified to distinguish documented mitigation, technical diagnosis, investigation outcomes or external intervention from a publicly reconstructable internal review and escalation pathway. No framework field, coding category or analytical methodology was changed. This dataset contains the cleaned supplementary evidence record supporting The AI Incident Actionability Gap: Public Evidence and Operational Interpretability in AI Governance. The workbook documents the structured qualitative coding of 18 publicly documented AI incidents across seven categories of operationally relevant information: (i) system identification, (ii) deployment context, (iii) timeline visibility, (iv) affected population, (v) mitigation visibility, (vi) evidence provenance, and (vii) oversight-route clarity. It includes case-level coding outcomes, supporting rationales, ambiguity notes, principal public evidence sources and evidence-environment metadata. The dataset supports research on AI governance, AI incident reporting and monitoring, post-deployment governance, public evidence, and operational interpretability.