Epistemological Sequencing as an Architectural Principle: Structural Output Properties of Multi-Regime Inference over Large Language Models

This open repository publication presents empirical evidence that the architectural structure of reasoning — rather than model capability or prompt sophistication — shapes the structural properties of outputs produced by large language models in analytical tasks. The study introduces and tests the principle of epistemological sequencing: the controlled succession of distinct inferential regimes, each with its own validity rules, admissible evidence, and inference operators, applied sequentially to the same analytical object. A controlled comparison was conducted across six analytical frameworks operating under epistemologically heterogeneous regimes — drawing on complex systems theory, cognitive neuroscience, phenomenology, narrative epistemology, adaptive governance, and morphogenetic dynamics. Each framework was tested in two conditions over the same organizational case (a regional healthcare network in transformation, anonymized): a sequenced execution following the framework's regime succession, and a single-prompt control generated independently via Google AI Studio under blind briefing. Four structural metrics were applied to the resulting outputs — M1 variable diversity, M2 relational density, M3 configurational differentiation, M4 inferential traceability — and three independent assessors evaluated the same outputs under blind conditions. The pre-specified success criterion (≥3/4 metrics favorable to the sequenced condition in ≥5/6 frameworks) was met in all six frameworks: four frameworks achieved 4/4 metrics favorable to the sequenced condition; two achieved 3/4 with a tie on M1. Relational density (M2) and inferential traceability (M4) were favorable to the sequenced condition without exception across the six frameworks. The blind evaluation confirmed the structural advantage in five of six frameworks by majority, with two by unanimity; one framework produced no majority verdict and is documented as an informative exception. The paper identifies what it terms epistemic monoculture — the structural convergence of analytical outputs across systems that rely on a single inferential regime — as an architectural rather than capability-level phenomenon, and argues that architectures that orchestrate distinct epistemological regimes preserve diversity of reasoning, relational structure, and audit traceability in ways that single-regime prompting does not.