Normative Frame

A normative frame is a deontic abstraction that constrains what an agent (or coalition) must, may, or must not do while leaving how to do it open. Calvanese, De Giacomo, Kampik, Lesperance, Marrella, Matta 2025 argue for elevating it to a first-class BPM abstraction alongside goals and operational specifications.

Normative vs operational

Operational specs (BPMN, DECLARE) describe procedures or allowed sets of activities — how to execute.
Normative frames describe obligations, permissions, prohibitions — what the outcome space must respect.

Classical BPM languages are entirely operational. DECLARE is declarative but still operational because it specifies sets of allowed activities, not deontic duties.

Why first-class?

With LLM-based agents enacting processes, sufficient autonomy for agents to synthesize their own plans requires decoupling the goal (what to achieve) and the frame (what must not be violated) from any single procedural script. Agents then synthesize operational strategies that are provably frame-compliant.

Challenges

Frame meta-model and hybrid specification languages (symbolic + subsymbolic, e.g. LLM extraction of frames from policy text).
Operationalization on real-world symbolic data at scale, with explainability.
Projection of global frames to local frames in distributed multi-agent scenarios.

concepts/framed-autonomy · concepts/agentic-bpm · concepts/explainability-apm · frameworks/bdi-agents