Asymmetric Intelligence: A Framework for Expectation Enforcement, Violation Dynamics, and Domain Viability Constraints

This paper advances asymmetric intelligence as a framework for reasoning about intelligence in enforcement governed environments, where the costs of error are discontinuous and survival depends on not crossing critical boundaries. In contrast to standard artificial intelligence systems optimized for average predictive success, the framework centers expectation, violation, regime change, and viability as the core terms of intelligence under consequence. It distinguishes descriptive measures from prescriptive survival constraints, introduces gated forms of adaptation bounded by invariants and viability conditions, and proposes a narrative architecture that ties human readable outputs to machine verifiable grounding. Together, these elements form a unified account of consequential intelligence that brings together insights from safe reinforcement learning, constrained optimization, runtime verification, change point detection, and grounded language generation.