Annotation Guidelines for LLM-Generated PRCT Classifications: Protocol for Human Auditing in Parliamentary Debates, Manifestos, and Web Archives

This document establishes the formal validation protocol used by human experts to audit and verify zero-shot Large Language Model (DeepSeek-v3) predictions concerning Population Replacement Conspiracy Theories (PRCT). Unlike traditional ab initio annotation frameworks, this protocol outlines a model-in-the-loop auditing workflow where human operators assign a binary decision (CONFIRM/REJECT) to the LLM's classification across long-form textual archives. The guidelines provide specific diagnostic criteria to police the boundary between generic anti-immigration rhetoric and structured conspiratorial frameworks within three distinct communicative domains:1. Parliamentary Debates (Institutional Register and Rhetorical Maquillage)2. Political Manifestos (Programmatic and Party Registers)3. Web Archives (Digital Fringe Registers) This methodological documentation ensures the reproducibility of the gold standard used to evaluate psycholinguistic and stylistic variation in radical political discourse within the framework of the HYBRIDS project.