RI–HITL: A Human-in-the-Loop Framework for Auditable AI-Assisted Qualitative Text Analysis

This working paper introduces the Recursive–Iterative Human-in-the-Loop framework (RI–HITL), a structured and auditable workflow for AI-assisted qualitative text analysis. The framework is designed to improve procedural consistency, transparency, traceability, and bounded procedural reproducibility while retaining human interpretive authority at all analytically consequential stages. RI–HITL combines corpus delimitation, relevance identification, segmentation, calibration, explicit decision rules, recursive–iterative coding, logged revision, saturation stress-testing, analytical freezing, and post-freeze synthesis. AI is treated as a bounded support system rather than an autonomous analytical judge: permissible uses include retrieval, orientation, consistency checking, clustering, and metadata support, whereas final coding, unreviewed theory-bearing reinterpretation, and autonomous inference remain non-permissible. The paper includes a worked example, operational templates, an extended AI-governance checklist, and an appendix specifying how RI–HITL may be used downstream of AST-derived theorem structures during empirical operationalisation and instantiation testing.