Architecture and Experimental Framework of the RHEA Cognitive Test Harness v3

Abstract The RHEA Cognitive Test Harness v3 is a structured experimental framework designed to evaluate intelligent agents operating in adversarial, ambiguous, and corrupted environments. Unlike conventional evaluation systems that measure only task accuracy or reward outcomes, the harness was designed to expose the internal dynamics of cognitive reasoning under hostile or deceptive conditions. This paper describes the architecture of the harness, the design principles used to construct the experimental environment, the structure of the task suite, and the evaluation pipeline used to generate the results presented in the companion papers. The harness introduces several key mechanisms including adversarial task generation, tactical correctness measurement, entropy monitoring, and recursive stabilization tracking. The framework was built to support recursive cognitive architectures such as RHEA‑UCM while remaining compatible with baseline heuristic systems. By instrumenting the reasoning process itself, the harness allows researchers to observe how agents respond to adversarial structures rather than merely whether they succeed or fail.