Temporally extended dissipative blankets: Boundary history as a non-Markovian screening mechanism

Biological boundaries are often modeled as instantaneous Markov blankets that separate internal and external states through sensory and active variables. However, real biological interfaces are dissipative systems with finite response times, relaxation, noise, and delayed transduction. Here, we examine whether such physical boundary dynamics can produce a non-Markovian screening relation, in which conditional separation is recovered only when boundary history is included. We constructed a minimal stochastic dissipative loop linking external, sensory, internal, and active states, with delayed sensory-to-internal coupling. Conditional mutual information was used to quantify residual internal-external dependence after conditioning on either the instantaneous sensory-active boundary or a temporally extended boundary containing finite sensory history. In the baseline linear model, instantaneous conditioning failed to fully screen internal from external states in a moderate fraction of simulations, whereas adding sensory history consistently reduced residual dependence and produced negative history-response trends. The effect was stronger in a nonlinear saturating model and was abolished when either external-to-sensory or sensory-to-internal coupling was removed. Sensitivity analyses showed that temporally extended screening depends on the balance among delay, dissipation, feedback, and noise, but is robust to changes in history-sampling stride. These results support the concept of a temporally extended dissipative blanket: a physical boundary mechanism whose statistical screening function is carried not only by its present state, but also by finite boundary history.