Focus Area: Referral routing, handoff logic, trust signaling, permissioned delegation, and agent-to-agent escalation paths used to pass users, tasks, or evidence to better-suited AI systems or human operators.
This ontology provides citation-quality definitions for 15 foundational terms, backed by authoritative sources from standards bodies (NIST, W3C, IETF, OASIS, ISO) and peer-reviewed research.
15
Technical Terms
75%+
Tier-1 Sources
V1.72
Pipeline Version
Technical Glossary
AGT001Referral Trigger Signal
marks the condition that causes an agent to stop handling a task directly and route it to a different system or human operator. It matters because referral quality starts with a clear, machine-recognizable reason for handoff rather than an ad hoc escape from uncertainty.
routes a task according to demonstrated competence, permission, and fit instead of simple availability or static routing tables. It matters because intelligent referral depends on matching work to verified capability, not just forwarding it somewhere else.
adjusts referral choices using identity, integrity, and prior performance signals from candidate recipients. It matters because a handoff is only useful when the receiving system is both capable and trustworthy for the exact class of task.
defines the rules that push a case from one service tier to another when risk, ambiguity, or authority requirements change. It matters because escalation should be governed by declared policy rather than hidden heuristics that are hard to audit.
captures the permission that allows one agent to share task context, user data, or evidence with another party for a bounded purpose. It matters because referrals often fail governance review when delegation is assumed rather than explicitly scoped.
bundles the facts, prior steps, and supporting artifacts that must travel with a referred case so the receiver can act without re-creating context. It matters because handoff quality collapses when the receiving party gets the task but not the evidentiary trail.
provides proof about the identity, authority, and role of the system accepting a referral. It matters because sending agents should verify who is taking custody before handing off sensitive tasks or context.
connects automated referral logic to a human decision-maker when policy, judgment, or accountability requires a person in the loop. It matters because not all referral endpoints are machines, but the bridge still needs structure and evidence continuity.
detects and blocks circular routing patterns in which tasks bounce among agents without converging on a responsible endpoint. It matters because referral systems become operationally toxic when escalation creates motion without progress.
records the moment responsibility for a case moves from one agent or operator to another. It matters because reliable referrals need a crisp transition point for accountability, timing, and audit.
defines how the receiving party sends status, decisions, or completions back to the original routing system. It matters because closed-loop referrals require not only handoff out, but reliable return of results and disposition.
assigns urgency and handling precedence to a referred case so downstream systems know how quickly and under what service expectations to respond. It matters because a handoff without priority semantics can misroute time-sensitive work into ordinary queues.
holds the structured state that should remain intact across a referral, including entities, prior decisions, and evidence references. It matters because useful handoffs preserve interpretive context instead of forcing the receiver to guess what the sender meant.
limits what the receiving party may do after a referral is accepted. It matters because forwarding a task should not silently grant unlimited authority over data, actions, or downstream delegation.
evaluates whether a prospective recipient is the right destination for a specific case based on role, trust, and outcome history. It matters because referral intelligence is strongest when it scores fitness rather than assuming one endpoint fits all exceptions.