A-IPA Comparator and Revision Replication Package for the Verify-or-Avoid Study

Version 1.3. This release replaces version 1.2, whose files are no longer distributed. Four files in version 1.2 carried working references to project-internal documents and to a file not included in the package; those references now point to material inside this package. No research content changed: the corpus, personas, prompts, rubric, responses, coding outputs, analyses, and human ratings are byte-identical. Public release of the file-bearing comparator and revision replication package for the manuscript: Shin, K. S., Yang, J., Min, Y., Kang, I. S., and Lee, M. (2026). Verify-or-Avoid: Accountability Concentration in AI-Mediated Senior Leadership: An Augmented IPA Study of Lived Accountability and Document-Grounded AI Reasoning. Manuscript under review. Contents. Five leader-anchored persona documents with a generic baseline and a twelve-section template; the 200-scenario stratified corpus with its expansion script; the four framework-condition prompts; the five-category advisory stance rubric; the substrate runner and automated-coding scripts; all 2,400 comparator responses; first-pass and second-pass automated stance labels; the captured statistical outputs; the deterministic revision analyses with a pinned environment; and the identity-minimised independent human provenance ratings for the 453 screened propositions, with the frozen overlap statistics and every retained disagreement. Reproduction. Two of the three revision scripts execute directly from this release. analysis_R1_documentary_support_audit_v2.py reproduces the 453-candidate screening universe and passes its fixture, corpus-regression, and arithmetic checks. analysis_R1_exact_pvalues.py matches all 40 reported manuscript assertions. A SHA-256 manifest covers every file. Withheld until publication. One file from the package supplied to the Editorial Office is not included here: a snapshot of the manuscript's Findings section, which is text under peer review. It serves only as the input to the persona firewall audit, whose captured output is retained in the package; that script detects the withheld input and reports it rather than failing opaquely. The snapshot will be added when the article is published. Not included. Interview transcripts and diary records are excluded for participant confidentiality. The persona documents are researcher-authored instruments anchored to the published participant profiles at the level of industry, role, organisational scale, approximate tenure band, accountability structure, and decision archetype only; names, organisations, locations, education histories, and narrative details are fictional, and they are not participant records. Version history. Version 1.0 (restricted) and version 1.1 (comparator materials only) are retained as separate versioned records and are not overwritten. The concept DOI 10.5281/zenodo.20200358 always resolves to the newest version.