Human Detection of AI-Generated Phishing: Study Protocol and Dataset Design for the Threat Terminal Experiment

v1.1: Appended protocol addendum documenting the removal of the email authentication header inspection panel from the platform interface. See Addendum section for rationale and impact on data collection. Original protocol text is unmodifiedThis paper presents the study protocol, dataset design, and methodological framework for an experiment examining which phishing techniques produce the lowest human detection rates when all stimuli are AI-generated at consistent linguistic quality. Data is collected through Threat Terminal, a purpose-built game-based research platform where participants classify AI-generated emails as phishing or legitimate, express confidence in each decision, and receive forensic signal breakdowns after each session. The dataset comprises 1,000 cards across six phishing technique categories and three legitimate email categories. Hypotheses are stated prior to full analysis. Initial pilot data from 70 participants and 996 classified cards confirms platform viability. Empirical findings will be reported in a separate publication upon reaching the target sample of 100 participants.