This paper develops a structural characterization of deterministic polynomial-time computation based on informational filtrations induced by execution. Building on earlier work showing that polynomial-time algorithms cannot eliminate global inconsistency at an exponential informational cost, it defines compatibility complexes and associates to each computation a monotone filtration reflecting progressive refinement of compatible inputs. The Informational/Filtrational Principle is proved, showing that such filtrations admit only locally certified changes and cannot create higher-dimensional global dependencies. Using this framework, the paper formalizes Global Non-Local Irreducible Dependencies and proves that they are incompatible with polynomial-time computation and preserved under standard reductions. As a consequence, the P versus NP problem is reduced to the question of whether NP-complete problems necessarily contain such dependencies.
Michael Arias (Wed,) studied this question.