DEEP SYSTEMISING AS EVOLUTIONARILY RETAINED COGNITIVE ARCHITECTURE: Causal-Depth Processing as a Qualitatively Distinct Architecture in High-Systemising Autistic Profiles Across Neurological, Evolutionary, Psychological, and Organisational Levels of

This paper proposes a cross-level theoretical account of high-systemising autistic cognition. The central claim is that autistic systemising does not simply occur more frequently than neurotypical systemising — it terminates at a different level. Neurotypical systemising stops at the surface pattern sufficient for social coordination; autistic causal-depth systemising continues to the underlying causal logic that makes the pattern necessary. That distinction — a difference of kind rather than degree — is the framework's central theoretical proposition. The argument develops across four levels of analysis: Neurological: reduced thalamic gating efficiency, arising from excitatory-inhibitory imbalance and aberrant precision weighting, produces a system in which prior beliefs are insufficiently weighted to halt processing at the template level. Evolutionary: specific Neanderthal-derived variants enriched in autistic individuals (Pauly et al., 2024), interpreted as consistent with a retained ancestral processing style maintained through de novo regeneration and episodic frequency-dependent selective value. Psychological: causal-depth processing generates characteristic costs (generative masking, rest-resistant burnout, HPA axis amplification) and outputs (first-principles problem-solving, structural identification of system failure). Sociological: the architecture is proposed to function as a structural catalyst at institutional change points, at characteristic personal cost. A processing-budget construct is developed to explain within-population variation in observable systemising, identifying a clinically significant, budget-depleted subpopulation that is indistinguishable from architectural non-systemisers on current instruments. Nine testable hypotheses (H1–H9) are generated, together with a four-phase empirical research programme gated on the development of a validated processing-depth instrument. The paper is a theoretical synthesis rather than an empirical study. All citations have been verified against primary sources; inferential extensions beyond the cited evidence are flagged explicitly throughout. This is a pre-print made available for scholarly engagement prior to peer review.