ATB 2.0 – Augmented Theory Building as Adversarial Theory Testing

Abstract / Description This paper introduces Augmented Theory Building (ATB) 2.0, a methodological framework for adversarial theory testing under AI-mediated research conditions. The increasing use of large language models in theoretical work has altered evaluative dynamics: coherence, fluency, and internal consistency can be rapidly produced without corresponding exposure to epistemic risk. Under conditions of high feedback and alignment, theories may appear robust while remaining insufficiently constrained. ATB addresses this problem by formalizing epistemic constraints designed to reintroduce resistance, loss, and scope limitation into theory evaluation. The method enforces a prohibition of affirmation, strict role asymmetry, context reduction, temporal distance, and—critically—a multi-model architecture to counteract time-dependent calibration effects inherent in prolonged human–AI interaction. ATB is a post-hoc discipline rather than a theory-construction method. Its primary output is not refinement or optimization, but negative epistemic yield: the identification of overextension, non-applicability, and justified reduction of theoretical claims. A longitudinal case study demonstrates how ATB narrows and hardens an existing theoretical framework through adversarial testing rather than improvement. The paper concludes by delineating the limits, non-applications, and risks of misuse of ATB. ATB does not make theories truer. It makes them less misleading