Key terms used in the BPM/Agent Stack specification. For Intent Stack terminology, see the Intent Stack glossary.

---

Activity — A unit of work within a governed process. In the BPM/Agent Stack, each activity carries up to 21 typed governance attributes specifying authorization, responsibility, inputs/outputs, system access, and governance documentation.

Activity-Level Governance Attributes — A set of 21 typed attributes attachable to any activity in a governed workflow. Derived from RACI, SIPOC, Value Stream Mapping, and ISO 31000. Include authorization constraints, system access scope, required inputs/outputs, escalation triggers, and governance documentation requirements.

BPMN 2.0 — Business Process Model and Notation, an OMG standard for graphically representing business processes. Provides the structural vocabulary for process decomposition, exception handling, escalation routing, and activity-level governance.

BPM/Agent Stack — The execution governance specification defining how authorized work gets done with process discipline. Three governance concerns: Orchestration, Integration, Execution.

Constitutional AI — Anthropic’s approach to training-time governance. The behavioral substrate beneath which no agent operates.

DMN 1.0 — Decision Model and Notation, an OMG standard for modeling and executing business decisions. Provides deterministic decision tables with defined hit policies.

Deterministic/Probabilistic Separation — The structural distinction between decisions that must be deterministic (compliance classification, escalation routing, boundary enforcement) and decisions that benefit from LLM judgment (observation, analysis, hypothesis generation).

Execution Governance — The per-activity controls that operate after identity and authorization are established. Specifies how authorized work gets done with process discipline, responsibility assignment, deterministic decision separation, structured exception handling, and audit trails.

FIRST Hit Policy — A DMN decision table evaluation mode where rules are evaluated in priority order and the first match wins. Used for escalation routing.

Governed Decomposition — Breaking a process into subprocesses, each with its own governance interface — authorization scope, acceptance criteria, exception handling, and audit trail.

Hit Policy — A DMN concept specifying how a decision table resolves when input matches rules. UNIQUE (exactly one match) for classification. FIRST (priority-ordered) for routing.

Intent Stack — The companion specification providing governance context: what an agent is authorized to do, by whom, under what constraints, and how alignment is assessed. Four concerns: Intent Discovery, Intent Formalization, Specification, Runtime Alignment. Published at intentstack.org.

Orchestration — How multiple agents are coordinated to execute specifications. One of the three execution governance concerns this specification addresses.

Integration — How governed agents connect to external systems with governed access scope. One of the three execution governance concerns.

Execution — How actual work is performed within governing constraints. One of the three execution governance concerns.

RACI — Responsibility assignment matrix: Responsible (does the work), Accountable (owns the outcome), Consulted (provides input), Informed (notified of outcome).

Stitching Mechanism — The bidirectional interface where the Intent Stack’s governance context meets the BPM/Agent Stack’s execution governance. Governance intent becomes actionable execution specification.

Structured Exception Handling — Typed exceptions with defined responses and escalation paths, as opposed to generic retry loops. Derived from BPMN’s event taxonomy.

Three-Layer Architecture — Constitutional AI (substrate) → Intent Stack (governance context) → BPM/Agent Stack (execution governance). Seven governance concerns across two companion specifications plus the Constitutional AI foundation.

UNIQUE Hit Policy — A DMN decision table evaluation mode where exactly one rule matches for any input. Used for unambiguous classification.