Intelligence Tests Measure Cognitive Style Conformity, Not Intelligence: A Two-Phase Association Test Framework for Revealing the Unmeasured Dimensions of Neurodivergent Cognition

Conventional intelligence tests operationalize intelligence as the speed and accuracy of convergentresponses to pre-defined problems—a cognitive performance characteristic of neurotypical processingthat constitutes one component of intelligent functioning, not its totality. This paper argues that themeasurement paradigm itself embeds four structural assumptions (efficient inhibitory control,hierarchical semantic processing, stable attentional focus, and task atomization) that privilege aspecific cognitive style, introducing construct-irrelevant variance correlated with neurodevelopmentalprofile. We demonstrate that composite IQ scores are mathematically designed to discard profiletopology—the most structurally informative feature of cognitive assessment—and that the g-factor,while capturing real shared variance, operates as a level parameter blind to the shape of individualcognitive architecture. We propose a two-phase association test framework—timed free association(Phase 1) followed by untimed metacognitive explanation (Phase 2)—grounded in dual-processtheory, that renders observable three cognitive dimensions structurally inaccessible under currenttesting: associative topology, metacognitive honesty, and System 1–System 2 translation capacity.Applying the framework to ADHD, autism spectrum conditions, and twice-exceptionality, we generatefalsifiable predictions and argue that what the convergent-speed paradigm registers as deficit is betterunderstood as cognitive signature: a distinctive pattern of cognitive organization rather than adegraded approximation of a neurotypical default. We further argue that the cognitive capacitiesmeasured by current tests are precisely those being rendered redundant by artificial intelligence,while the capacities the tests ignore—problem structuring, associative creativity, metacognitiveself-regulation, and resistance to fluency-mediated epistemic deception—are those that retainirreplaceable human value. The framework is explicitly distinguished from Gardner's multipleintelligences theory: we do not propose new types of intelligence, do not challenge the g-factor, andprovide falsifiable predictions. This is a theoretical paper; empirical validation constitutes thesubsequent research program.