Mission Control Glossary

Target architecture — not the current operating contract

This glossary includes future product terms. For current roles, resources, and procedures, see the current release.

In the current release, a caller-owned W3C Trace is the top-level operational correlation boundary and an optional namespace-scoped Session groups Team activity. Mission Control exposes no Run resource or Trace/Session lifecycle API. Definitions that name a Run below describe target architecture only.

This glossary is the canonical vocabulary for product, design, engineering, and operating documents.

Platform terms

Agent — A registered probabilistic actor with a durable identity, owner, environment-scoped grants, policy bindings, keys/workload identity, Runs, and Traces. It may run in a future Mission Control native runtime or an external runtime.

Agent Team — A versioned composition of agents with declared specialist roles, delegation and handoff rules, shared limits, and termination conditions.

Deployment — A managed release of a resource into an environment, used by product areas such as Model Management and potentially a future Agent Builder. Current external Agent execution does not create this resource; observed runtime version fields are not attestation.

Control Kernel — The mandatory platform foundation that owns shared identity, registry, lifecycle, policy, approval, evidence, observability, evaluation, and configuration semantics.

Module — A composable product capability that uses Control Kernel contracts for its resources and operations. Modules add focused authoring or delivery behavior without redefining shared control semantics.

Governed Run — A run tied to a registered principal, resource, immutable version, environment, policy, trace, and evidence record, in which all model and MCP calls use their enforced gateways. A run that bypasses a required gateway is not governed.

Model Gateway — The exclusive OpenAI-compatible inference boundary that resolves virtual models to approved deployments, isolates provider credentials, enforces access and routing controls, and emits usage, cost, trace, and evidence data.

Connector Gateway — The exclusive invocation boundary for registered MCP tools. It resolves tool versions, validates typed input and output, obtains access and policy decisions, manages approval pauses, invokes the tool, and publishes evidence.

MCP Server — A registered endpoint that exposes typed enterprise tools through the Model Context Protocol. Server identity, ownership, transport, authentication reference, environments, and health are governed independently of its tools.

Connector — Mission Control's internal product abstraction over one discovered MCP tool. Each connector has its own identity, owner, version, primary class, risk posture, capability grants, certification state, health, usage, and lifecycle.

Virtual Model — A stable logical model identity used by agents and workflows. A routing policy resolves it to an approved model deployment without changing the consuming resource's identity or history.

Workflow — A versioned durable process of typed steps and explicit transitions across agents, connectors, people, timers, branches, transformations, and sub-workflows. Automations owns workflow state.

AI Application — A governed user experience bound to registered agents and workflows, published as a responsive web, embedded, conversational, launch, task, result, or evidence surface.

Retrieval Collection — An owned, purpose-specific grouping of approved Data connector sources with access, retention, freshness, indexing, citation, and evaluation rules.

Connector classifications

Every connector has exactly one primary classification. Classification is assigned to the individual MCP tool, not its server.

Data Connector — Reads, searches, lists, fetches, or supplies information without changing enterprise state. It is side-effect-free from the caller's perspective and is governed for source access, data classification, purpose, lineage, freshness, and citation.

Decision Connector — Evaluates, scores, recommends, validates, or applies a business rule without changing enterprise state. It is side-effect-free from the caller's perspective and records input and output references, decision version, confidence or validation signals, and outcome linkage.

Action Connector — Changes enterprise state, emits an externally meaningful event, or initiates work that can cause a side effect. It requires strong authorization, explicit scope, risk and value controls, a stable idempotency key, approval where required, operating-window rules, and recovery or compensation expectations.

Any externally meaningful side effect makes a tool an Action Connector, regardless of the tool or server's label. Composite behavior uses a Workflow containing multiple connectors rather than weakening this classification.

Execution and evidence terms

Policy Decision — The versioned result of evaluating a principal, resource, operation, context, and relevant history. Its outcome is exactly allow, deny, or approval_required, with an explicit reason.

Governance Approval — A durable policy outcome that routes a specific request to authorized approvers with context, reason, separation-of-duties rules, rationale, delegation, escalation, and expiry. Current identity, policy, version, and context are re-evaluated before execution resumes.

Human Task — A durable Workflow step in which a person performs business work. It may share an inbox with Governance Approvals but has a distinct identity, permission model, evidence trail, and resumption contract.

Evidence Event — An append-only, attributed record of a control change, decision, execution fact, evaluation, incident, correction, or outcome using the shared event envelope.

Run — One top-level execution instance initiated by a user, event, API, application, agent, or workflow. It provides the correlation boundary for resources, decisions, calls, costs, states, evaluations, and outcomes.

Trace — The causally ordered spans and events within a Run, including application interactions, agent activity, workflow steps, model calls, connector calls, policy decisions, approvals, and errors.

Outcome — A referenced business or operational result linked to a Run, action, evaluation, or incident. An Outcome is distinct from successful technical completion.

System of Precedent — The reconstructable chain connecting goal and actor, resource identities and versions, business context, policy and authority, model and connector evidence, exceptions and approvals, executed actions, evaluations, and outcomes. It is structured business and operating evidence, not hidden model reasoning.

Current governance mapping

The current source release uses Organizations, Teams, real user/service-account virtual keys and separate human management roles. Agent is managed application content; deployments own secretless execution access and publisher funding. Observe uses Requests, Traces, Sessions and Usage with separate metadata/content/export permissions. Historical identity terms in this target proposal are not the operating contract. See Current release.