Meaning, Opacity, and Function: An Observer-Relative Classification of Artifacts

Every working epistemic system – a research group, a software stack, a model with its operators – holds artifacts it uses without understanding: models invoked through interfaces, cryptographic routines run as black boxes, legacy code executed daily. Whether an artifact is understood, merely apprehended, or functional is not a property of the artifact alone but of the observer that confronts it. We take that observer's situation as the origin. From a single capacity – to understand, or to fail to understand, what one apprehends – two semantic predicates differentiate: structured meaning (M) and residual opacity (C); the independent capacity to act supplies a third axis, functionality (K). Fixing these three capacities as independent predicates over an artifact domain yields seven provably exhaustive regions – one of which an explicit convention stipulates uninhabited for the single-observer classification, and which is recovered per observer once the predicates are relativized; relativizing them to each observer makes region membership provably non-invariant across agents; and their overlap – the domain of structured ambiguity Θ = M ∩ C – is the natural habitat of symbols and denotation. The framework is deliberately static and extensional – no possible worlds, no accessibility relations, no temporal structure – and its proven core stays close to its definitions: a partition theorem, the predicate-toggle symmetry it carries, non-invariance across observers, and the combinatorial and compositional facts these determine. Read at the level of a single agent, that partition yields a classification of the qualitative types an agent's three extensions can take when all three are non-empty and functionality is read objectively – 109 of them, or 34 up to relabelling of the predicates, against 52 labelled types under the subjective reading – and, when such a state is drawn as three disks in the plane, two of those types are realized by no three distinct circles in any position whatever, one of the two being a limit of the plane rather than of the drawing conventions. A further layer is static-generative: the functional predicate K is itself generated by composition from a primitive basis, stratified by derivation order rather than execution time, and this compositional genesis is proved to be a conservative extension of the static classification. An execution layer, equally static, relates region assignments by a step relation carrying no time index and no cost: of the eight predicate toggles, exactly one non-identity toggle is coherent from every region an agent can occupy, so admissibility is a property of the (source, toggle) pair and not of the toggle alone. Internal generation is proved to reach its fixed point at the compositional closure of the basis and to admit no further generation step from there; the relation holds the agent's architecture fixed throughout, and therefore models no acquisition. Its contribution is conceptual: a natural, first-principles vocabulary for the semantic–functional status of artifacts – observer-relative and explicit about the configurations it excludes – that offers a way to conceptualize complex processes in terms of what an agent understands, merely operates, or cannot yet reach, and thereby lays groundwork for articulating an artificial agent's semantic and functional states in philosophically explicit rather than purely operational terms. A closing appendix carries a single non-operational subject, by way of an argument in the natural-theology genre, through the whole apparatus – the placement in the symbol domain of the statements that carry it, the two readings of K, non-invariance across observers, and the strata lying outside the horizon – and exhibits a placement that moves with the observer rather than with the subject; no result of the paper depends on it. Region dynamics and the geometry of the functionality axis are developed in companion papers.