8. The Activity Model

(Normative)

The fundamental unit of work in the BPM/Agent Stack is the governed activity — a unit of agent or human execution carrying a full set of governance attributes. This contrasts with the fundamental units in current agent frameworks, which carry only mechanism attributes (prompt, tools, data/state).

A conformant Activity Model implementation SHALL carry all 21 attributes defined in §8.1 through §8.5. Attributes MAY have null values where not applicable to a specific activity instance, but the attribute structure SHALL be present.

8.1 Role Attributes (derived from RACI)

Attribute	RACI Mapping	Agent Application
Participant	Responsible	Agent role or human actor executing this activity
Accountable Owner	Accountable	The human(s) answerable when this execution step is questioned. SHALL always be a human in governed agent deployment.
Consulted	Consulted	Subject matter experts, specialized agents, or knowledge sources consulted during execution
Informed	Informed	Stakeholders, downstream actors, governance observers notified of completion or exception

8.2 Data Lineage Attributes (derived from SIPOC)

Attribute	SIPOC Mapping	Agent Application
Suppliers	Supplier	Upstream process, agent, or human providing input data. SHALL establish explicit provenance.
Inputs	Input	Typed, vocabulary-controlled data entering the activity. SHALL be explicit data contracts, not unstructured context injection.
Outputs	Output	Typed, vocabulary-controlled data produced by the activity. SHALL be explicit deliverables, not unstructured LLM responses.
Customers	Customer	Downstream process, agent, or human consuming outputs. SHALL establish delivery accountability.
Systems	Process infrastructure	MCP servers, APIs, tools, and external services the agent invokes during this activity. SHALL be typed, governed, and auditable.

8.3 Performance Attributes (derived from Value Stream Mapping)

Attribute	VSM Concept	Agent Application
Cost	Process cost per unit	Token cost, API cost, compute cost, external service cost per execution of this activity
Work Time	Processing time	Active agent processing time
Wait Time	Queue/delay time	Queue time, dependency wait, human-in-the-loop latency, external API response time
Total Time	Lead time	End-to-end duration including work time, wait time, and overhead
Value-Add	Value-add vs. waste	Does this step add value to the end deliverable, or is it overhead, rework, or waste?

Performance attributes SHOULD be captured for every governed activity execution. Implementations SHOULD support aggregation of performance attributes across process instances to enable Value Stream Analysis.

8.4 Risk Attributes (derived from ISO 31000)

Attribute	ISO 31000 Concept	Agent Application
Risk	Risk identification and assessment	Failure modes, likelihood, severity, and controls in place at this step
Problems	Known issues / active risks	Active issues, edge cases, or failure patterns currently affecting this activity

Risk attributes SHALL be assessed for each governed activity. Activities with high-severity risk ratings SHALL have documented mitigation controls.

8.5 Governance and Documentation Attributes

Attribute	Source Practice	Agent Application
Documentation	BPM CBOK §2 (Process Modeling)	Step-level execution instructions — the “prompt” in agent context, distinguished from governance constraints and decision logic
Attachments	Reference materials	Templates, examples, few-shot examples, grounding documents available during execution
Policy Links	Compliance/governance practice	Linked regulatory, corporate, or operational governance documents accessible at point of execution
Comments	BPM CBOK §3-4 (Analysis/Design)	Review notes, governance observations, design rationale, and process improvement annotations
Custom Fields	Domain extension	Organization-defined attributes for industry-specific governance requirements

8.6 Activity Types (per BPMN 2.0)

Each governed activity SHALL be typed according to the BPMN 2.0 activity taxonomy:

Activity Type	BPMN 2.0 Definition	Agent Mapping
User Task	Requires human action	Human-in-the-loop step: review, approval, judgment call
Service Task	Automated system execution	Agent tool invocation, API call, MCP server interaction
Business Rule Task	Executes decision logic	DMN decision table evaluation — deterministic, not LLM-inferred
Script Task	Executes code	Agent code execution
Send Task	Dispatches a message	Agent sends output to another agent, human, or system
Receive Task	Waits for a message	Agent blocks until input arrives from external source
Manual Task	Physical-world action	Human performs work outside any system
Subprocess	Governed decomposition	Sub-agent execution with own process model, boundaries, and governance interface