De-actorhood and the Redistribution of Actorhood: Documentary Dataset and Codebook from the Ex-GKN Factory Struggle, 2018–2026

This dataset accompanies the article “Filling the Void: De-actorhood and the Redistribution of Actorhood in the Ex-GKN Factory Struggle.” It contains the documentary evidence and coding structure used to analyse the 2018–2026 dispute surrounding the former GKN plant in Campi Bisenzio, Italy. The analytical corpus comprises 211 documents, of which 201 generated coded material and 10 were consulted but yielded no passage meeting the coding criteria. The final dataset contains 574 coded attribution events. The workbook includes four main analytical components: a Source Register listing all 211 documents in the analytical corpus, with source identifiers, titles, authors or issuing bodies, publication dates, document types, provenance information, access status, coding status, and coded-record counts; an article-aligned Codebook containing 31 first-order codes, organised into five second-order families and three aggregate dimensions: Withdrawal, Transfer of the burden, and Re-anchoring; Coded Data containing the 574 retained attribution-event records, including source identifiers, event and document dates, acting and addressed entities, original excerpts, analytical paraphrases, primary and secondary codes, code families, aggregate dimensions, provenance, source location, and evidentiary confidence; a Source Reconciliation register documenting seven discrepancies among sources and explaining how they were treated analytically. The unit of analysis is the attribution event, defined as a passage in which an identifiable party attributes or withholds action, responsibility, capacity, obligation, or status. Coding followed an abductive, theory-informed approach. All coding was conducted manually by a single author. Coding decisions were checked against the underlying source material, earlier records were reviewed after the codebook was frozen, and a second corpus-wide verification pass was conducted. Evidentiary assessment was kept distinct from coding. Claims used to support analytical mechanisms were evaluated by provenance and corroboration, and mechanism claims required support from more than one provenance class. Discrepancies among sources were retained as analytically relevant rather than silently harmonised. Each coded record also includes an evidentiary confidence classification: A — High / direct: the attribution is explicitly stated in the source; M — Medium / indirect: the attribution is supported indirectly by a documented sequence or cross-source discrepancy; B — Low / secondary: the claim depends on a secondary account of an unavailable document or on an assertion not independently verified in the underlying source. Confidence refers to the explicitness and documentary support of the attribution, not to the truth of the underlying claim. The dataset is primarily based on Italian-language documents. Original source titles, excerpts, actor labels, and analytical notes are retained in Italian where appropriate, while the coding architecture and methodological metadata are provided in English. Source files themselves are not redistributed where copyright or access restrictions apply. The workbook instead provides bibliographic and source-location information to support transparency and traceability. Claude (Anthropic) and ChatGPT (OpenAI) were used only for language checking, editorial refinement, and other non-analytical consistency checks. They were not used for coding, codebook development, data interpretation, or theoretical analysis.