FPF Reference

The Agential Role & Agency Spectrum

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

“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

  • agency as role
  • agency spectrum
  • contextual role assignment
  • autonomy grading
  • substrate-neutral autonomy.

Relations

A.13builds onRole Taxonomy
A.13builds onU.RoleAssignment: Contextual Role Assignment
A.13builds onExternal Transformer & Reflexive Split (C-2)
A.13informsAgency‑CHR
A.13informsRoC‑Autonomy Budget & Enforcement
A.13builds onHolonic Foundation: Entity → Holon
A.13coordinates withMeta-Holon Transition (MHT): Recognizing Emergence and Re-identifying Wholes
A.13coordinates withTrust & Assurance Calculus (F–G–R with Congruence)
A.13coordinates withMulti-Scale Ethics Framework
A.13coordinates withAgency‑CHR
A.13explicit referenceTransformer Constitution (Quartet)
A.13explicit referenceExternal Transformer & Reflexive Split (C-2)
A.13explicit referenceHolonic Foundation: Entity → Holon
A.13explicit referenceAdvanced Mereology: Components, Portions, Aspects & Phases
A.13explicit referenceMeta-Holon Transition (MHT): Recognizing Emergence and Re-identifying Wholes

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 upon the foundations laid in the FPF Kernel to provide that definition. A.1 established that only a U.System can be the bearer (holder) of behavioral roles. A.2.1 defined the universal U.RoleAssignment (Holder#Role:Context) as the canonical way to assign roles. A.3 and A.12 defined the TransformerRole and the principle of the external agent.

The intent of this pattern is to:

  1. Formally define agency not as an intrinsic type of holon, but as a contextual Role Assignment.
  2. Introduce a measurable, multi-dimensional spectrum of agency via a dedicated Characterization (Agency-CHR), 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 level of autonomy 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 artifacts (U.Episteme), 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 U.Agent as a new base type alongside U.System and U.Episteme would violate Ontological Parsimony (C-5) and create conflicts with the dynamic nature of roles. A system might act as an agent in one context and a passive component in another; a static type cannot capture this.
  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

ForceTension
Scientific Fidelity vs. SimplicityContemporary science (e.g., Active Inference) models agency as a continuous, scale-free spectrum. FPF needs to honor this rigor while providing a simple, teachable model for practitioners.
Role vs. TypeThe intuition is to think of an "Agent" as a type of thing. FPF's architecture demands that it be modeled as a role to preserve dynamism and ontological hygiene.
Measurement vs. LabelEngineers and managers need a quick, intuitive label (e.g., "this is a Level 3 agent"), while formal assurance requires a detailed, multi-dimensional, evidence-backed measurement.
System-only Action vs. Collective ActionHow does agency apply to groups like teams or swarms? This requires a clear link to the rule from A.1 that any acting group must be modeled as a U.System.

Solution

FPF's solution is threefold: it defines an Agent via U.RoleAssignment (A.2.1), makes agency measurable with a dedicated Characterization, and provides a didactic summary via a graded scale.

The Core Definition: Agent as a Contextual Role Assignment

An "Agent" in FPF is not a fundamental type. It is a convenience term (a Register 1 / Register 2 label) for a specific kind of Contextual Role Assignment (U.RoleAssignment):

Agent ≍ U.RoleAssignment(holder: U.System, role: U.AgentialRole, context: U.BoundedContext)

This means an Agent is simply a U.System that is currently playing an AgentialRole within a specific U.BoundedContext.

  • No U.Agent Type: To be clear, there is no U.Agent base type in the FPF Kernel. This avoids type inflation and preserves the dynamic nature of roles.
  • Epistemes Cannot Be Agents: As the holder must be a U.System, this definition constitutionally forbids U.Epistemes from being agents, preventing the "episteme-as-actor" category error.
  • Canonical Syntax: The technical notation for an agent is System#AgentialRole:Context.

The AgentialRole and its Specializations

  • U.AgentialRole: This is the abstract U.Role that grants a U.System the capacity for goal-directed action within a context. It is the "license to act."
  • Specialized Roles: More specific behavioral roles like TransformerRole and ObserverRole are considered specializations or sub-roles of AgentialRole. They describe what kind of agential action is being performed at a given moment.
    • A system playing TransformerRole is an Agent that is currently modifying another holon.
    • A system playing ObserverRole is an Agent that is currently gathering information. This creates a clean hierarchy: a Transformer is always an Agent, but an Agent is not always a Transformer (it could be observing, planning, or idle).

Measuring Agency: The Agency-CHR and the Spectrum

Agency is not a binary switch; it is a multi-dimensional spectrum of capabilities. FPF models this using a dedicated pattern, Agency-CHR (C.9), which is a Characterization that attaches a set of measurable properties to a U.RoleAssignment.

The Agency-CHR profile is grounded in contemporary research (e.g., Active Inference, Basal Cognition) and includes the following key characteristics. Each is measured for a specific agent in a specific context and must be backed by evidence (A.10).

  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 agent'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 agent can pursue, from simple set-points to abstract, multi-scale goals.

Context-bounded task-family specialization claims

When work shifts to a new TaskFamily, describe the holder as acquiring context-bounded task-family specialization rather than as becoming more generally intelligent in the abstract. The same holder may carry different task-family specializations across different task families without becoming a new kernel type. Breadth across unrelated task families is not the governed claim here; the governed claim is time-to-usable specialization on the declared task family and work target under a named work-measure threshold, adaptation budget, and freshness or provenance basis.

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-CHR 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 CHR profile into an intuitive level of autonomy.

GradeLabelTypical Agency-CHR Profile (Conservative Lower Bound)Archetypal Example
0Non-AgentialBMC ≈ 0, PH ≈ 0, MP ≈ 0A rock, a document, a passive structural component.
1ReactiveBMC > 0, PH ≈ 0, MP ≈ 0A thermostat; a simple feedback controller. Follows fixed rules.
2PredictiveBMC > 0, PH > 0, MP ≈ 0A model-predictive controller with a fixed model; a chess engine that plans moves but doesn't learn new strategies.
3AdaptiveBMC > 0, PH > 0, MP > 0A self-calibrating sensor system; a machine learning agent that updates its model with new data.
4Reflective/StrategicHigh BMC, PH, MP, PER, and OC. Capable of meta-cognition (reasoning about its own reasoning) and pursuing abstract goals.An autonomous R&D system; a cohesive, self-organizing DevOps team.

Crucial Distinction: The Agency-CHR profile is the normative evidence. The Grade is a pedagogical shortcut. An artifact cannot claim a grade without having a corresponding, auditable CHR profile to back it up.

Archetypal Grounding

The universal pattern of agency, defined as a Contextual Role Assignment and measured by the Agency-CHR, manifests across all domains. The following table demonstrates its application to the FPF's two primary archetypes: a U.System and a collective U.System (a team), while explicitly showing why a U.Episteme cannot be an agent.

ArchetypeHolder (U.System)Role & Context (#Role:Context)Agency-CHR Profile SketchResulting Agency Grade
Simple ControllerThermostat_Model_T800#AgentialRole:HomeHeatingSystemBMC: High (maintains temp).
PH: Zero (no prediction).
MP: Zero (fixed logic).
PER: Very High.
OC: Low (single set-point).		Grade 1 (Reactive)
Advanced ControllerPredictiveCruiseControl_v3#AgentialRole:VehicleDynamicsBMC: High.
PH: High (predicts traffic flow).
MP: Zero (fixed model).
PER: High.
OC: Medium (optimization).		Grade 2 (Predictive)
Learning SystemSelfCalibratingSensorArray#AgentialRole:IndustrialProcessBMC: High.
PH: High.
MP: Medium (learns drift).
PER: High.
OC: Medium.		Grade 3 (Adaptive)
Collective AgentDevOpsTeam_Phoenix (a collective U.System)#AgentialRole:ProjectPhoenixBMC: High (maintains velocity).
PH: High (sprint planning).
MP: High (retrospectives).
PER: Medium-High.
OC: High (abstract business goals).		Grade 4 (Reflective/Strategic)
Knowledge ArtifactISO_26262_Standard.pdf (U.Episteme)N/A (Cannot be a holder of an AgentialRole)N/AGrade 0 (Non-Agential)

Key takeaway from grounding: This table makes the abstract model concrete. It shows that the FPF agency model can precisely differentiate between simple controllers and complex learning systems. It also reinforces the Strict Distinction principle: the ISO standard (U.Episteme) is a crucial justification (justification?) for the actions of an agent (like the DevOps team), but it is never an agent itself.

Conformance Checklist

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

IDRequirement (Normative Predicate)Purpose / Rationale
CC-A13.1 (Holder Type)The holder of a U.RoleAssignment with role: U.AgentialRole MUST be a U.System.Prevents the "episteme-as-actor" category error. Enforces Strict Distinction (A.7).
CC-A13.2 (RoleAssignment Mandate)Any claim of agency MUST be represented by a complete U.RoleAssignment instance, including an explicit holder, role, and context.Ensures that agency is always modeled as contextual and bound to a specific bearer, not as a free-floating property.
CC-A13.3 (CHR Evidence)Any claim about an Agent's grade or level of autonomy MUST be substantiated by an auditable Agency-CHR profile with Evidence Graph Ref (A.10).Makes claims of agency falsifiable and prevents "agency by marketing."
CC-A13.4 (Grade is Didactic)The Agency Grade (0-4) SHALL NOT be used as a normative input for formal reasoning. It is a didactic summary of the Agency-CHR profile.Prevents oversimplification in formal models. The detailed profile, not the summary grade, must be used for assurance cases.
CC-A13.5 (Collective as System)To claim agency for a collective (e.g., a team, a swarm), the collective MUST first be modeled as a U.System with a defined U.Boundary and a coordination U.Method.Prevents the error of assigning agency to a mere set or collection (MemberOf). Aligns with A.1 and A.14.
CC-A13.6 (MHT for Emergent Agency)If a collection of systems, previously non-agential or at a lower grade, develops a new supervisory structure and crosses a documented Agency-CHR threshold, a Meta-Holon Transition (MHT, B.2) MUST be declared.Makes the emergence of collective agency an explicit, auditable event, preventing "magic" emergence.

Consequences

BenefitsTrade-offs / Mitigations
Category Safety & Clarity: The pattern provides a clear, unambiguous definition of agency that prevents common modeling errors and is consistent across all of FPF.Increased Modeling Granularity: Requires modelers to think in terms of Role-assignments and contexts, which is slightly more complex than just labeling something an "Agent." Mitigation: The Holon#Role:Context syntax and tooling support make this manageable in practice.
Falsifiable & Measurable Agency: By grounding agency in the Agency-CHR, the framework transforms a vague philosophical concept into a set of concrete, evidence-backed engineering properties.Measurement Effort: Populating the Agency-CHR profile requires real work (testing, analysis, data gathering). Mitigation: The profile can be built iteratively. An initial estimate can be used, with the understanding that its Reliability (R) score is low until backed by evidence.
Scalable Autonomy Model: The graded scale provides a sophisticated language for describing and comparing different levels of autonomy, from simple automation to strategic intelligence.Risk of Misinterpreting Grades: The simple 0-4 scale could be misused as a simplistic marketing label. Mitigation: The normative requirement (CC-A13.4) to always link a grade to its underlying CHR profile acts as a guardrail against this.
Elegant Handling of Collectives: The pattern provides a clean way to model the agency of teams, swarms, and organizations without violating ontological principles.-

Rationale

This pattern's strength 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-CHR provides a direct, practical implementation of these ideas.
  • Ontologically Sound: By defining an Agent as a Contextual Role Assignment, the pattern avoids the ontological pitfalls of creating a new base type. It fully embraces the FPF's core architectural principle of separating substance (holder) from function (role) within a context. This aligns with best practices from foundational ontologies (like UFO) and the principles of Strict Distinction (A.7).
  • Pragmatic and Actionable: The pattern is designed for engineers and managers. The Agency Grade provides a quick communication tool, while the underlying Agency-CHR 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 Role Taxonomy: Provides the universal Contextual Role Assignment (U.RoleAssignment) mechanism.
    • A.12 External Transformer: The actions of an Agent are modeled using the external transformer principle.
  • Coordinates with:
    • B.2 Meta-Holon Transition (MHT): A significant jump in the Agency-CHR of a collective can trigger an MHT.
    • B.3 Trust & Assurance Calculus: The Agency-CHR profile provides crucial inputs for assessing the reliability and safety of an autonomous system.
    • D.2 Multi-Scale Ethics Framework: The Agency Grade is used to determine the level of moral responsibility and accountability assigned to a system.
  • Instantiates:
    • The Agency-CHR (C.9), which provides the formal definitions for the characteristics (BMC, PH, etc.).

A.13:End