Research paper 32

Operational Graphs for Agentic Work: Context, Evidence, Authority, and Verification

Agentic work is often described as prompts, loops, workflows, or model sessions. Those terms are useful, but they do not explain how a reviewer later knows what context was used, which evidence supported an action, which authority boundary applied, or what verification attached to the result. This paper proposes the operational graph as a missing middle object for agentic systems. An operational graph is a small, typed, inspectable relation model over work objects: tasks, runs, artifacts, approvals, checks, memory records, evidence packets, source paths, and proof receipts. Its purpose is not to be a universal ontology or a decorative visualization. Its purpose is to make agent work traversable and auditable. The local evidence comes primarily from Switchboard's work and context graph surfaces, Caliper's commit-pinned repository graph and telemetry-only measurement contract, and adjacent Research Papers manuscripts on proof-carrying development, authority contracts, governed harnesses, and advisory metrics. The strongest supported claim is a methods claim: graph-shaped operating records can help separate context assembly, evidence support, authority gating, and verification state. The manuscript does not claim that graphs improve agent quality, that graph orchestration outperforms linear workflows, or that any local graph grants release, legal, publication, or deployment authority.