Discoverable Agent Knowledge -- A Formal Framework for Agentic KG Affordances (Extended Version)

A new formal framework proposes a semantic layer to help autonomous agents determine what they can reliably prove from a knowledge graph, addressing a gap left by current metadata standards like VoID and DCAT [1]. The framework revisits a challenge posed to the Semantic Web Services community two decades ago: how agents with different ontological commitments could discover and use web services coherently [2]. That earlier work produced OWL-S and WSMO, formally grounded capability descriptions that specified what a service could do and what an agent needed to know for invocation to be epistemically sound [2]. Terry Payne and colleagues now extend those insights to the knowledge graph setting, arguing that existing metadata standards describe what a KG contains but say nothing about what a specific agent can prove from it [1]. A central concern is that in deployed KGs, the governing schema description logic and the operative entailment regime can diverge, creating an epistemic failure mode invisible to current metadata [2]. The authors propose a four-dimensional formal framework to address this: Semantic Expressivity, Agentic Discoverability, Task-Relative Grounding, and Epistemic Trust Scope [1]. From these dimensions they derive the Agentic Affordance Profile, or AAP, a semantic layer above VoID and DCAT intended to enable principled KG selection, composition, and failure diagnosis at agent planning time [2]. The concept of an affordance, borrowed from ecological psychology, refers to what an environment offers an organism. In embodied cognition research, cognitive functions are understood to be shaped by the bodily state and capacities of the organism, including its interactions with the environment [3]. The AAP framework operationalises a similar idea at the individual-agent level, specifying what a knowledge graph affords a particular agent given its task vocabulary and reasoning capabilities [1]. The paper includes a worked example drawn from a scholarly-search task to ground the framework concretely [1]. It also lays out a five-point research agenda identifying the formal, computational, and engineering work required to realise AAP-based affordance matching at scale [2]. The work appears as an extended preprint on arXiv, submitted on 18 May 2026 and revised on 25 May 2026 [1].