This paper proposes a theoretical framework for introducing deterministic bit constraints into large language model (LLM) inference spaces through recursive IF-THEN masking dynamics. The study extends previous work concerning deterministic extraction of prime distributions within binary computational space and defines the MRST (Multi-Record Structured Thread) protocol, which applies XOR interference structures and carry propagation dynamics to transform probabilistic token generation into logically convergent structures. In this framework, carry propagation from lower-order bits toward higher-order bits is interpreted as an irreversible bit-linked constraint chain. The study investigates whether such recursive constraints may reduce probabilistic instability inside LLM inference processes. Furthermore, the paper proposes that recursive masking structures may have theoretical implications for the analysis of carry-transition structures inside large-prime cryptographic systems such as RSA and ECC. This work presents these ideas as a structural interpretation model and requests independent mathematical and engineering verification regarding their validity in AI inference control and cryptographic analysis.
Minoru Yoshitake (2026) studied this question.