Distinguishing Two Kinds of Absence in Assertion Stores: Missing Instances versus Missing Representational Categories

Description : Systems that store and reason over assertions routinely encounter absence: a value is not present, a fact is not recorded, a query returns nothing. Most systems collapse these situations into a single undifferentiated state ("unknown", "null", "not found"), which conflates two epistemically distinct conditions. This paper argues for an explicit, first-class distinction between two kinds of absence: (A1) the absence of an instance within a category the system already represents, and (A2) the absence of any category capable of representing the situation at hand. We propose that a system should expose the first as actionable information and declare the second as a limit of its own representational scope, rather than silently forcing it into an existing category. This shifts a classic database distinction (Codd, 1986) from the applicability of a known attribute to the existence of a representational category. We specify this distinction, describe a minimal reference implementation demonstrating that it can be expressed as a general rule (not a per-case heuristic), and report an adversarial evaluation of that rule. We deliberately scope our claims: we show that the distinction is implementable and produces observably different behavior from a closed-world baseline that does not make it; we do not claim to have demonstrated that this behavior is preferable under an independent metric, which we identify as future work. We also report a dependency uncovered by adversarial testing: the reliability of the A1/A2 distinction is contingent on reliable matching between input terms and the system's schema, a separate and unsolved problem.