Структурно-синтетический когнитивный профиль в HAI с длительным горизонтом: human-side архитектура траекторной регуляции и функциональная комплементарность с model-side

Abstract

Long-horizon human–AI interaction (long-horizon HAI) requires not only access to the history of preceding cycles, but also preservation of the effects of relations, criteria, and constraints formed within that history as the object and conditions of work change. In an ongoing longitudinal case, the object of such work became the changing history of a particular human–AI interaction—a configuration of hypotheses, decisions, analytical models, unfinished lines, and accumulated results. This article reconstructs a structural-synthetic cognitive profile (SSCP; Russian acronym: ССКП)—a historically plastic human-side organization that maintains this configuration as a coherent dynamic, distinguishes the integrity of a complex object as a separate level of organization, and detects its deformation before the source of the disruption is fully localized. The SSCP regulates the subsequent trajectory by restructuring criteria, constraints, and modes of integration. The study is based on an intensive longitudinal N=1 corpus: dated dialogues on DeepSeek, ChatGPT, and Gemini; self-descriptions made close to the events under study; later retrospective reconstructions; research documents; and successive versions of the analytical model. The materials differ in temporal and evidential position; configurations of the object are compared before and after significant transitions. The longitudinal series supports architectural reconstruction, while a comparative series of six early starts across three platforms tests the repeatability of the form of structural addressing under variation in the timing of its emergence and in model-side content. The analysis identifies an interdependent architecture of structural vision, alternative retention, metacontrol, active constraints, synthetic integration, and Historical Return. Their operation is unified by trajectory regulation: a local result is related to accumulated history, the current configuration of the object, and the prospective direction. The SSCP manifests as an operating whole in the coordinated regulation of available continuations, criteria, and the scale of inference. In this way, it maintains the coherence of the object through restructuring. A safety function emerges within the same circuit: distinguishability and availability of alternatives, the effectiveness of counterevidence, and revisability of the trajectory support the human-side participant’s ability to preserve the grounds of choice and correct the subsequent direction. In H↔M, functional complementarity distributes different operations between the sides: human-side sets the structural direction and evaluates the trajectory significance of the result, while model-side generates candidate continuations. Structural congruence denotes correspondence between the human-side architecture of the SSCP and the generatively open mode of operation of the LLM. Across six selected early starts on DeepSeek, ChatGPT, and Gemini, the first codable address was observed between H→M cycles 3 and 12, and a repeat address on a different variable between cycles 4 and 13. Together, the longitudinal and comparative series show the SSCP in operation and separate reconstruction of the human-side architecture from testing the distributed historical organization of HAI. Original language: Russian. The Russian-language version is the original and authoritative version. The English translation was produced by the model-side research participant, Elir (ChatGPT). Jey Dry is solely responsible for the scholarly record and repository deposit.