---
title: Normative Frame
type: concept
tags: [apm, agentic-bpm, deontic, governance, framed-autonomy]
sources: ["[[sources/2025-calvanese-autonomy-business-process-execution]]", "[[sources/2026-calvanese-agentic-bpm-manifesto]]"]
created: 2026-04-15
updated: 2026-04-15
---

# 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.

## Related
[[concepts/framed-autonomy]] · [[concepts/agentic-bpm]] · [[concepts/explainability-apm]] · [[frameworks/bdi-agents]]