#The Agential Role & Agency Spectrum

A.13 · Stable · Part A - Kernel Architecture Cluster

#About this pattern

This is a generated FPF pattern page projected from the published FPF source. It is canonical FPF content for this ID; it is not a FPF Reference product feature page.

#How to use this pattern

Read the ID, status, type, and normativity first. Use the content for exact wording, the relations for adjacent concepts, and citations to keep active work grounded without pasting the whole specification.

“Agency is not a kind of thing; it is a way some systems operate.”

The concept of "agency"—the capacity of an entity to act purposefully—is central to engineering, biology, and AI, yet it remains one of the most overloaded and ambiguous terms. Without a precise, falsifiable, and substrate-neutral definition, models of autonomous systems risk descending into "self-magic," where actions have no clear cause and accountability is lost.

#Keywords

• exact System
• local agential system-role kind and criterion
• classification
• obtaining assignment
• scope
• working situation
• window
• evidence-backed core
• conditional characteristic profile
• autonomy grading.

#Relations

#Content

#Intent & Context

The concept of "agency"—the capacity of an entity to act purposefully—is central to engineering, biology, and AI, yet it remains one of the most overloaded and ambiguous terms. Without a precise, falsifiable, and substrate-neutral definition, models of autonomous systems risk descending into "self-magic," where actions have no clear cause and accountability is lost.

This pattern builds directly on the FPF Kernel. A.1 establishes that the acting holder must be a U.System. A.2 and A.2.1 distinguish a local system-role kind, classification by that kind, and an obtaining occurrence of a directly declared U.SystemRoleAssignment species. A.12 supplies the acting-side externalization principle.

The intent of this pattern is to:

1. Define agency not as an intrinsic type of holon, but as an assignment claim: an exact local agential system-role kind is assigned to a U.System through an obtaining U.SystemRoleAssignment.
2. Introduce a measurable, multi-dimensional spectrum of agency via a dedicated agency-characteristic profile, moving beyond a simple binary "agent/not-agent" switch.
3. Provide a clear, didactic grading system that allows engineers and managers to assess and communicate the Agency Grade of any system in a consistent, evidence-backed manner.

#Problem

If agency is treated as a monolithic, intrinsic property or a mere label, four critical failure modes emerge, undermining the rigor of FPF:

1. Episteme-as-Actor: Models might incorrectly assign agency to knowledge epistemes (U.Episteme) or their publications, leading to nonsensical claims like "the specification decided to update the system." This is a direct violation of Strict Distinction (A.7).
2. Type Inflation: Introducing a root agent kind alongside U.System and U.Episteme would violate Ontological Parsimony (C-5). The same System may qualify for an agential system-role kind and receive an assignment in one working situation but not another. The agency claim states its scope and window separately; a root type cannot express these differences.
3. Unfalsifiable Claims: Without a measurable basis, "agency" becomes a subjective label. A team might call their system an "agent" for marketing purposes, but this claim has no verifiable meaning and cannot be audited, violating Evidence Graph Referring (A.10).
4. The Binary Trap: A simple "agent/not-agent" classification is too coarse. It fails to distinguish between a simple thermostat, a predictive cruise control system, and a strategic, self-learning robotic swarm, even though their cognitive capabilities differ by orders of magnitude.

#Forces

#Solution

FPF's solution is threefold: establish agential participation through an obtaining system-role assignment, measure agency with a dedicated Characterization, and provide a didactic summary through a graded scale.

#The Core Definition: Agential participation through an exact system-role assignment

An ordinary-language "agent" is not a fundamental FPF type. When a precise agency claim is needed, name four things:

1. the acting holder recognized as a U.System;
2. the exact local agential system-role kind whose membership criterion the holder satisfies;
3. an occurrence of a directly declared U.SystemRoleAssignment species that assigns that kind to the holder and actually obtains; and
4. any claim scope, working situation, and time window needed by the intended use, kept separate from the assignment's identity.

This keeps a useful ordinary word without creating a universal Agent or AgentialRole kind. Classification by the local kind does not by itself establish an assignment or performed Work. Because the holder must be a U.System, an episteme cannot become the acting holder of this assignment.

#Local Agential System-Role Kinds and Their Specializations

• Local agential system-role kind: A practice or source may define a local kind whose stable work-facing contribution is goal-directed action. The kind classifies candidate Systems under its own criterion; it is not a universal root kind, an assignment occurrence, or Work.
• Specialized agential system-role kinds: A local practice may distinguish transformation, observation, planning, or another contribution when it supplies a real criterion for the distinction. An assignment to one such kind establishes only that assignment; any transformation, observation, plan, or performed Work still needs its own claim.

#Measuring Agency: The Agency Characteristic Profile and the Spectrum

Agency is not a binary switch; it is a multi-dimensional spectrum of capabilities. A.13 defines the current domain profile and attaches its measurable characteristics to the exact holder and agency claim; A.17, A.18, A.19, C.16, and A.10 govern characterization, measurement, and evidence. Planned C.9 Agency Characteristic Profile may later consolidate that profile but supplies no current definitions or governing force.

The agency-characteristic profile is grounded in contemporary research (e.g., Active Inference, Basal Cognition) and includes the following key characteristics. Each measurement names its exact holder and, where relevant, its task family or work target, claim scope, working situation, and time window; A.10 governs the claim-bound evidence and provenance account.

1. Boundary Maintenance Capacity (BMC): The ability of the system to maintain its structural and functional integrity against perturbations. (How robust is it?)
2. Predictive Horizon (PH): The temporal or causal depth of the holder's internal model. (How far ahead can it "see"?)
3. Model Plasticity (MP): The rate at which the agent can update its internal model (U.GenerativeModel) in response to prediction errors (U.Error). (How quickly can it learn?)
4. Policy Enactment Reliability (PER): The probability that the agent will successfully execute its chosen U.Method under operational conditions. (How reliably does it do what it decides to do?)
5. Objective Complexity (OC): A measure of the complexity of the U.Objective the holder can pursue, from simple set-points to abstract, multi-scale goals.

#Task-family specialization claims

When Work shifts to a new TaskFamily, describe evidence-backed specialization for that task family and work target rather than greater intelligence in general. Keep the task family, work target, claim scope, working situation, measurement window, work-measure threshold, adaptation budget, and provenance basis as separate values. The same holder may show different specializations for different task families without becoming a new U-kind; the claim here is time-to-usable specialization for the stated task family and target.

Low-human-overlap or newly discovered task families remain admissible when the task family, evidence basis, and reuse window are explicit by value.

#The Agency Grade (Didactic Layer)

While the multi-dimensional agency-characteristic profile is essential for formal assurance, engineers and managers need a simpler, at-a-glance summary. The Agency Grade is a non-normative, didactic scale from 0 to 4 that synthesizes the profile into an intuitive autonomy grade.

Crucial Distinction: The agency-characteristic profile is the normative evidence. The Grade is a pedagogical shortcut. A holder cannot claim an Agency Grade without having a corresponding, auditable characteristic profile to back it up.

#Archetypal Grounding

The cases below apply the same test to individual and collective Systems, and then contrast them with a knowledge artifact. Each positive case names the holder, one illustrative local agential system-role kind, and one distinct assignment occurrence that relates that holder to that kind. The characteristic sketch and grade remain separate claims. The names are didactic examples, not a universal AgentialRole vocabulary.

Key takeaway from grounding: The same ontology works for a thermostat, a predictive controller, a learning System, and a collective System: classification by a local kind and an obtaining assignment are both stated, while scope, situation, Work, evidence, profile, and grade remain separate. An exact ISO claim episteme may be cited in an A.10 evidence-provenance account or a B.3 reliance claim; any direct relation relied on by that account or claim must actually obtain under its own defining rule. Its file carrier merely bears a publication form. Neither the citation nor the publication acts.

#Conformance Checklist

To ensure the agency model is applied rigorously and consistently, all FPF publications must adhere to the following normative checks.

#Consequences

#Rationale

This pattern's value comes from its synthesis of contemporary, post-2015 research into a single, operational model.

• Grounded in Science: The move away from a binary, type-based view of agency towards a graded, spectrum-based model is directly aligned with modern research in Active Inference (Friston et al.), Basal Cognition (Fields, Levin), and evolutionary cybernetics. The agency-characteristic profile provides a direct, practical implementation of these ideas.
• Ontologically Sound: Agential participation uses an exact local system-role kind and a separately obtaining U.SystemRoleAssignment instead of a new base type. Holder, kind, classification, assignment, performed Work, claim scope, and time window remain distinct. This follows Strict Distinction (A.7) without making the practitioner carry a universal context tuple.
• Pragmatic and Actionable: The pattern is designed for engineers and managers. The Agency Grade provides a quick communication tool, while the underlying agency-characteristic profile provides the detailed, auditable data needed for formal assurance and risk management. This duality satisfies both Didactic Primacy (P-2) and Pragmatic Utility (P-7).

In essence, this pattern does not invent a new theory of agency. It distills and operationalizes the emerging scientific consensus, packaging it into a rigorous, falsifiable, and practical tool for the FPF ecosystem.

#Relations

• Builds on:
  • A.1 Holonic Foundation: Establishes that only U.Systems can be bearers of behavioral roles.
  • A.2 System-Role Kinds and Assignments: Distinguishes an exact local system-role kind, classification by that kind, and an obtaining U.SystemRoleAssignment.
  • A.12 Acting-Side Externalization and Reflexive Split: Work by an acting holder is modeled using the acting-side externalization principle.
• Coordinates with:
  • B.2 Meta-Holon Transition (MHT): A significant jump in the agency-characteristic profile of a collective can trigger an MHT.
  • B.3 Trust & Assurance Calculus: The agency-characteristic profile provides crucial inputs for assessing the reliability and safety of an autonomous system.
  • D.2 Multilevel Ethics For Holon Work: The Agency Grade is used to determine the moral-responsibility posture and accountability assigned to a system.
• Future consolidation:
  • Planned C.9 Agency Characteristic Profile may later consolidate the characteristics (BMC, PH, etc.), but provides no current definitions or governing force; the current profile is defined here and measured under A.17/A.18/A.19/C.16/A.10.

#A.13:End

---

Last Updated: 2026-09-10 — upstream FPF commit a87d0ef4 (github.com/ailev/FPF)