Appendix A. LLM-Based Concept Detection: Prompt Development and Classification Cases

This supplementary appendix documents iterative prompt development for large language model-based detection of three interpretive concepts in oral history interviews: collective memory, intersubjectivity through affirmation and backchanneling, and identity performance. Six tables present prompt modifications, their design rationales, selected classification results, and diagnostically selected interview passages illustrating true positives, false positives and false negatives. The tables examine the effects of operational definitions, conversational context, passage boundaries and relational criteria on classification, including trade-offs between precision and recall. The reported results document prompt development rather than independent evaluation. Examples from additional interviews illustrate recurring error types and the challenges of translating interpretive concepts into reproducible annotation decisions. Funded by the Deutsche Forschungsgemeinschaft (DFG) and the Arts and Humanities Research Council (AHRC) through “Mixed-Methods Digital Oral History: Interlocking Semantic Web Technologies and Historical Interpretive Analysis to better understand narratives of emergence, disruption and change in the History of Computer Technology in the Humanities”. DFG project number: 528052031.