A.13 - The Agential Role & Agency Spectrum
Status: Stable
A.13:0 - Use This When
Use this pattern when a precise claim of agency matters: which System qualifies for a local goal-directed system-role kind, and which assignment actually holds for the action in question? Start with that System, the kind and criterion it satisfies, the obtaining assignment, and only the scope, situation and window the use needs. Keep the evidence for those facts recoverable. The first result is a supported agency claim, not a five-characteristic measurement exercise.
Open the characteristic profile when a Grade, autonomy or profile claim consumes it, when the local kind’s criterion depends on a characteristic, or when a named assurance use requires it. Otherwise stop at the core. A characteristic-only claim about an admitted System need not assert a system-role assignment.
Ordinary use of “agent” does not by itself require this precise account. Use A.12 for an acting/changed-participant question and A.15.1 for whether a dated Work occurrence took place.
A.13:1 - Intent & Context
A.13 separates a precise agency claim from a claim about the holder’s agency characteristics. The former identifies a System’s local kind and obtaining assignment; the latter states the characteristics the receiving use needs.
A.1 establishes the holder as 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 helps recover the acting and changed participants when that distinction matters.
This distinction serves three uses:
- Define agency as an assignment claim: a local agential system-role kind is assigned to a
U.Systemthrough an obtainingU.SystemRoleAssignment, rather than being an intrinsic holon type. - Describe a multi-dimensional spectrum of agency through the agency-characteristic profile when its values are needed.
- Provide a didactic grading system for communicating a profile summary.
A.13:2 - Problem
Treating agency as an intrinsic property or a mere label hides different questions:
- Episteme-as-Actor: When “the specification decided to update the system” hides the actor, use A.12 to identify the proposed acting entity. A precise agency claim needs that entity admitted as a System under A.1, distinct from the specification under A.7.
- Type Inflation: Adding a root agent kind duplicates existing ontology when System, local kind and assignment already express the needed claim. The same System may qualify for an agential system-role kind and receive an assignment in one working situation but not another. Scope and window remain separate; a root label alone does not express these differences. A.11 tests a proposed ontology addition for material loss under the best existing expression.
- Unsupported Claims: A label such as “agent” supplies neither the local kind criterion and obtaining assignment nor evidence for a claimed autonomy profile. Those claims need their own support under A.10.
- The Binary Trap: “Agent/not-agent” alone does not distinguish a thermostat’s feedback, a cruise-control system’s prediction and a strategic robotic swarm’s self-learning capabilities.
A.13:3 - Forces
| Force | Tension |
|---|---|
| Scientific Fidelity vs. Simplicity | Spectrum accounts draw attention to different capacities and levels of agency. A teachable summary still needs to state which characteristics and measurement bases it uses. |
| Role vs. Type | The same System can qualify for a local agential kind and hold its assignment in one situation but not another. Characteristic claims are additional when consumed by the use. |
| Measurement vs. Label | A quick label such as “Level 3 agent” summarizes more detailed claims. When assurance consumes agency characteristics, it needs their evidence-backed values, not that label. |
| Collective Agency | A precise agency claim for a team or swarm needs the collective admitted as a U.System, not merely named as a group. |
A.13:4 - Solution
Establish the agency core when making a precise agency claim. Build the characteristic profile when the receiving claim requires it; use a Grade only as its didactic summary.
A.13:4.1 - 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:
- the acting holder recognized as a
U.System; - the exact local agential system-role kind whose membership criterion the holder satisfies;
- an occurrence of a directly declared
U.SystemRoleAssignmentspecies that assigns that kind to the holder and actually obtains; and - 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 establish an assignment or performed Work. An episteme cannot be this assignment’s holder because the holder must be a U.System. Support each core fact under its own criterion, using characteristic results when that criterion requires them. A characteristic-only claim about an admitted System need not add an assignment.
A.13:4.2 - 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 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.
A.13:4.3 - Measuring Agency: The Agency Characteristic Profile and the Spectrum
The agency-characteristic profile describes distinct capabilities of the holder. Use it for a consumed Grade/autonomy/profile claim, a characteristic-dependent local criterion or a named assurance use that requires it. A.13 defines this domain profile; A.17, A.18, A.19, C.16 and A.10 govern characterization, measurement and evidence.
The following descriptions state what the five characteristics concern. A measured value also needs its declared characteristic definition, scale, measurement method and evidence basis under the patterns above. Each measurement names its holder and, where relevant, task family or work target, claim scope, situation and time window. A.10 governs the claim’s evidence and provenance.
- Boundary Maintenance Capacity (BMC): The ability of the System to maintain its own structural and functional integrity against perturbations. The maintained property is the holder’s integrity, not merely the value it controls.
- Predictive Horizon (PH): The temporal or causal depth of the holder’s internal model. State which depth the claim uses; the horizon is distinct from prediction accuracy.
- Model Plasticity (MP): The rate at which the holder can update its internal model in response to prediction errors. Describe the model update and its rate; evaluate capability acquisition separately under E.10.LRN.
- Policy Enactment Reliability (PER): The probability that the holder will successfully execute its chosen
U.Methodunder operational conditions. The event is successful Method execution; a result or safety claim needs its own basis. - Objective Complexity (OC): A measure of the complexity of the objective the holder can pursue, from simple set-points to abstract, multi-scale goals. These examples distinguish objectives; a complexity comparison needs a declared criterion and scale.
Build the profile for its receiving use. First name the claim or decision that needs a characteristic. Select the characteristics whose values can change that answer. Keep the holder, configuration, task or Method, assistance and relevant conditions common across the results you combine. Reuse a sufficient existing account; a characteristic-only claim can remain separate from an assignment claim.
For each selected question, use or construct a domain definition that connects an observable distinction to a characteristic value. A.17 and A.18 separate the characteristic, its scale and allowed operations from preference or scoring. C.16 supplies measurement discipline. Explain how the observation distinguishes the values needed by the receiving use. If the same observations fit materially different values, obtain a discriminating observation or retain that uncertainty.
| Characteristic | Construct the measurement needed by the use |
|---|---|
| BMC | Identify the holder’s structural and functional properties to preserve, the perturbations, tolerated loss and recovery, and external assistance. Test that integrity condition on the specified cases. Report the perturbations withstood, a supported limit or a success proportion for a declared population. |
| PH | Name the internal model and choose temporal or causal depth. Establish the future interval or kind and number of causal transitions that the model represents. Report that extent in its declared units; assess prediction accuracy separately when the receiving use needs it. |
| MP | Identify the changing model, prediction-error signal, measure of model change and time or budget denominator. For a scalar parameter, state whether the result uses net change or total variation; these differ when updates oscillate. Interpret the achieved rate under its observed conditions. |
| PER | Identify the chosen Method, successful-execution event, operational population and probability model. Include required operations and permitted stops in the event definition. A sample frequency estimates a probability only under the stated sampling and inference assumptions. |
| OC | Describe the objective and select a complexity criterion, such as independently adjustable targets or a declared rubric for coordinating conditional goals. Establish that the holder can pursue an objective with that structure. A demonstrated objective can supply a result without establishing the holder’s maximum capacity. |
Retain each value or admissible value range, uncertainty, evidence and applicability. An unmeasured characteristic remains unmeasured. Evidence that a holder has no predictive mechanism can establish zero PH; missing information about that mechanism cannot. A model-update rate and a task-performance improvement answer separate questions. A.13:4.3.1 connects a supported adaptation result to the task-family claim.
Assemble only the results that the receiving use consumes, and apply its comparison or decision rule. Values require compatible meanings and conditions, or a justified transformation, before comparison. A faster model change can be useful correction or harmful oscillation; the receiving objective determines that preference. Use the Grade mapping in §4.4 only when a didactic summary is wanted. After a change, reopen the dependent result: new measurement error changes its warranted value range; a tighter task tolerance can change suitability while leaving that range intact; a new model can require a different PH or MP definition.
A.13:4.3.1 - Task-family specialization claims
When assessing a holder’s time-to-usable specialization, name the holder, TaskFamily and work target. Use C.22 for the task anchor and C.22.1 for adaptation time or budget to the declared work-measure threshold. Keep task family, work target, claim scope, situation, measurement window, threshold, adaptation budget and provenance basis distinct where the claim uses them. The same holder can show different specializations for different task families; this does not establish greater intelligence in general or a new U-kind.
For a task-family comparison described as low-human-overlap, use C.22:5.2.1 to state the human-practice reference set, correspondence and coverage rule. Newly discovered families retain the same explicit task-family, evidence and reuse-window conditions. Such a comparison characterizes the task-family relation; the holder’s specialization still needs its own result.
A.13:4.4 - The Agency Grade (Didactic Layer)
Engineers and managers can use the Agency Grade as a non-normative, didactic summary of a profile. The scale runs from 0 to 4; an assurance use that consumes agency characteristics needs their supported values, not this summary.
| Grade | Label | Illustrative agency-characteristic profile | Archetypal Example |
|---|---|---|---|
| 0 | Non-Agential | BMC ≈ 0, PH ≈ 0, MP ≈ 0 | An admitted System whose supported profile has the stated values. A document or another non-System has no Agency Grade. |
| 1 | Reactive | BMC > 0, PH ≈ 0, MP ≈ 0 | A thermostat; a simple feedback controller. Follows fixed rules. |
| 2 | Predictive | BMC > 0, PH > 0, MP ≈ 0 | A model-predictive controller with a fixed model; a chess engine that plans moves but doesn’t learn new strategies. |
| 3 | Adaptive | BMC > 0, PH > 0, MP > 0 | A self-calibrating sensor system; a machine learning agent that updates its model with new data. |
| 4 | Reflective/Strategic | High 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. |
The Grade 0 label summarizes a profile; it does not settle the separate local-kind classification or assignment claim.
The profile states characteristic claims and their supporting evidence; the Grade is a pedagogical summary. An actual Grade claim needs a supported profile and a stated mapping from its values to the label, including thresholds when the mapping uses them. Without that basis, leave the Grade unresolved. CC-A13.3 governs that support; CC-A13.4 excludes the Grade as a normative premise.
A local mapping may cover only the profiles needed for its communication use. State non-overlapping conditions for the labels it assigns and leave other cases unclassified. When a supported value is a range, check the permitted values against those conditions. One label is supported only when all consequential admissible values select it; otherwise retain the uncertainty or alternatives allowed by the mapping. Preserve the characteristic results that the label does not summarize.
A.13:5 - Archetypal Grounding
Start with the controller: “Thermostat_Model_T800 holds T800-home-heating-assignment as HomeHeatingController for household-temperature control.” In this constructed case, Thermostat_Model_T800 identifies one installed unit, admitted as a System, in House-H1 from 08:00 to 12:00 on Day-D.
The local HomeHeatingController criterion is the ability to read the household temperature and selected set-point, command heating below the set-point and stop heating at or above it. The stipulated device behavior satisfies that criterion at the selected 21 °C set-point. The directly declared HomeHeatingAssignment species assigns a qualifying System to that kind when the household controller designation and enabled control connection obtain; both hold for T800 in that household and window.
The core result is that this System qualifies for HomeHeatingController and holds the obtaining assignment for the stated use. No Grade or autonomy measurement is needed for that result. If the household withdraws the designation while the device retains its capabilities, the assignment ceases under its rule; classification may remain true, and the precise agency claim must change. The table gives further schematic variations for individual and collective Systems, then a knowledge-artifact contrast. Its local kind names and profile/Grade assumptions are illustrative.
| Archetype | Holder (U.System) | Illustrative local agential system-role kind | Distinct obtaining assignment occurrence | Agency-characteristic profile sketch (illustrative assumptions) | Illustrative Agency Grade |
|---|---|---|---|---|---|
| Simple Controller | Thermostat_Model_T800 | HomeHeatingController | T800-home-heating-assignment assigns Thermostat_Model_T800 to HomeHeatingController for household-temperature control. | BMC: High (assumed). PH: Zero (no prediction). MP: Zero (fixed logic). PER: Very High. OC: Low (single set-point). | Grade 1 (Reactive) |
| Advanced Controller | PredictiveCruiseControl_v3 | VehicleDynamicsController | PCC-v3-vehicle-dynamics-assignment assigns PredictiveCruiseControl_v3 to VehicleDynamicsController for the driving situation supplied by the working case. | BMC: High. PH: High (predicts traffic flow). MP: Zero (fixed model). PER: High. OC: Medium (optimization). | Grade 2 (Predictive) |
| Learning System | SelfCalibratingSensorArray | IndustrialProcessAdaptiveController | sensor-array-process-adaptation-assignment assigns SelfCalibratingSensorArray to IndustrialProcessAdaptiveController for the calibration task family and window supplied by the working case. | BMC: High. PH: High. MP: Medium (assumed). PER: High. OC: Medium. | Grade 3 (Adaptive) |
| Collective acting holder | DevOpsTeam_Phoenix (a collective U.System) | ProjectPhoenixDeliveryCoordinator | phoenix-team-delivery-assignment assigns the collective System DevOpsTeam_Phoenix to ProjectPhoenixDeliveryCoordinator for the project work being claimed. | BMC: High (assumed). PH: High (release planning). MP: High (assumed). PER: Medium-High. OC: High (abstract business goals). | Grade 4 (Reflective/Strategic) |
| Knowledge artifact | No acting holder. ISO_26262_Standard.pdf is a file carrier; the selected standard edition and any exact claim episteme made available through it remain distinct. | N/A | N/A: neither the carrier nor an episteme is a U.System, so neither can receive an agential system-role assignment. | N/A | N/A — no System holder or agency-characteristic profile |
The profile values above are illustrative assumptions, not measured results. The case explanations support narrower statements:
- The thermostat “maintains temperature” in the household-control use. That does not establish the thermostat’s own structural and functional integrity under perturbation;
BMC: Highremains an assumption. - “Learns drift” leaves the sensor’s model/parameter-update versus capability-acquisition meaning unspecified.
MP: Mediumdoes not supply that missing account. - The team’s retrospectives name an activity, not a measured rate of internal-model update.
MP: Highremains an assumption; E.10.LRN keeps model change and capability acquisition separate.
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 assurance claim; any direct relation relied on by that account or claim must actually obtain under its own defining rule. The file carrier bears an expression conforming to a publication form.
A.13:5.1 - Construct a measured profile for a calibration use
Consider one installed SensorArray-S7, admitted as a System. This constructed case supplies its local IndustrialProcessAdaptiveController criterion and obtaining assignment: S7 detects deviation from a reference signal, selects a permitted offset update or stops calibration on an invalid input, and has the designation and enabled connection for that control. Those premises establish the agency core. The additional question is what its bench profile shows and whether calibration reaches residual error at most 0.1 °C within three minutes.
The measurement definitions and observations below are stipulated case inputs. The internal sensor model is z = r + b, where z is the raw indication, r the reference value and b the estimated offset. The reference relation establishes a constant raw offset z - r = 0.6 °C for this test. The admitted OffsetUpdateMethod-3 reads that reference, checks the allowed update and commits the new offset once per minute:
b_next = b + 0.5 * (z - r - b).
Starting at b = 0, the three committed offsets are 0.3, 0.45 and 0.525 °C. The measurement basis bounds the total uncertainty in interpreting residual error by ±0.02 °C. After the third update, that residual therefore lies in [0.055, 0.095] °C, satisfying the 0.1 °C target. The following profile keeps the other supported characteristics separately interpretable.
| Result | Basis and use boundary |
|---|---|
| BMC: integrity maintained in four of four specified perturbation cases | Integrity I requires an intact executable, one valid configuration and restoration of both channels within five seconds. The supplied test histories establish I after each one-second interruption of power, one channel or communication, and after an interrupted parameter write. Recovery is automatic after the interruption ends. This is coverage of those four cases. |
| PH: 30 seconds temporal depth | S7’s internal forecasting function produces z_hat(t+h) = r(t) + b(t) for 0 <= h <= 30 s, holding the current reference and offset over that prediction interval. The implementation supplies that model extent. Prediction accuracy is a separate question. |
| MP: 0.175 °C per minute over the three updates | The declared measure is total variation of the stored offset divided by elapsed time. All three updates have the same sign, so both total variation and net change equal 0.525 °C; divide by three minutes. Parameter readback resolution is 0.01 °C. |
| PER: 0.95 for uniform selection from the stated finite population | The population is 20 fully examined scripted bench episodes with valid inputs. Successful execution means completing the required reads, checks and commits of OffsetUpdateMethod-3. Nineteen histories meet it; one records a failed commit. Under uniform selection of one of these episodes, the probability is 19/20. A production-population estimate would need its own sampling and inference basis. |
| OC: one independently adjustable target pursued in this case | The local complexity criterion counts independently adjustable target quantities. Here the target quantity is residual temperature error; the three-minute budget and offset-update bounds are fixed conditions of this comparison. The calculation establishes pursuit of this objective in the stated regime. |
For the C.22.1 adaptation question, the first successful minute checkpoint is at 180 seconds, with a budget of three reference/update cycles and the stated initial model. This threshold result uses the residual-error bound; MP reports the parameter-change rate.
For a local didactic convention, assign Grade 3 within this controller family when BMC covers the named perturbation set, temporal PH is positive and error-driven MP is distinguishable from zero at the declared resolution. Leave other profiles unclassified. S7 receives that label; PER and OC remain explicit profile results outside what this partial mapping summarizes.
Changed receiving condition. Tighten the residual-error target to 0.05 °C while keeping the three-update budget. The profile values and local Grade remain supported, but the calibration result fails the new target: even the lower residual bound after three updates is 0.055 °C.
Under the same model, five updates would give an upper residual bound of 0.03875 °C. The practitioner can compare a larger adaptation budget with another update Method. The changed tolerance reopens the adaptation decision.
A.13:6 - Conformance Checklist
Apply these normative checks to the corresponding agency, profile or Grade claim in an FPF publication:
| ID | Requirement (Normative Predicate) | Purpose / Rationale |
|---|---|---|
| CC-A13.1 (Holder Type) | The holder System of an obtaining agential U.SystemRoleAssignment MUST be a U.System. | Prevents the “episteme-as-actor” category error. Enforces Strict Distinction (A.7). |
| CC-A13.2 (Assignment Mandate) | A precise claim of agency MUST name the exact local agential system-role kind and an obtaining occurrence of a directly declared U.SystemRoleAssignment species. Any claim scope, working situation, and time window needed by the use remain separate. | Binds agency to a specific holder and assignment without turning a generic context field into their identity. |
| CC-A13.3 (Characteristic Evidence) | An actual claim about a holder’s Agency Grade or autonomy profile MUST have a supported agency-characteristic profile with recoverable evidence and provenance under A.10. A Grade claim also states the mapping from those values to its didactic label. Use an Evidence Graph Ref when that representation is needed. | Supports the characteristic values and the Grade interpretation. |
| 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-characteristic profile. | An assurance case that consumes these characteristics uses their supported values, not the Grade. |
| CC-A13.5 (Collective as System) | To claim agency for a collective (e.g., a team or swarm), it MUST first be admitted as a U.System under the complete A.1 criterion. | Distinguishes an admitted collective System from a mere set or collection (MemberOf); parthood claims use A.14. |
| CC-A13.6 (MHT for Emergent Agency) | If new supervision and a documented agency-characteristic threshold crossing raise a whole-reidentification question about an established prior whole, apply B.2’s comparison. When B.2 establishes the transition, a Meta-Holon Transition (MHT) MUST be declared. | First constitution or recognition without an established prior whole uses A.1 and the applicable construction and kind-specific patterns. If the same whole explains the changed facts, stop without MHT. A prior description as non-agential or at a lower didactic Grade may describe the starting case; it is not a normative premise. |
A.13:7 - Consequences
| Benefits | Trade-offs / Mitigations |
|---|---|
| Agency core: Local kind, classification and obtaining assignment answer different questions; the same System may qualify and be assigned in one situation but not another. | Modeling granularity: Distinguish those claims and any performed Work; state scope or window when it changes the claim. Use a short ordinary-language account first and expose identifiers only when needed. |
| Characteristic claims: The profile separates the capabilities being claimed and the evidence needed for their values. | Measurement effort: Testing, analysis and data gathering take work. Build the profile iteratively; an initial estimate can be used with its uncertainty and evidence limits explicit. Missing support for that estimate does not imply low System reliability. |
| Didactic communication: The 0–4 scale gives a short vocabulary for discussing profiles, from simple automation to strategic intelligence. | Grade misuse: A Grade claim needs the underlying profile under CC-A13.3 and cannot replace its characteristic values in formal reasoning under CC-A13.4. |
| Collective agency: The same core distinguishes agency claims for admitted team, swarm and organization Systems. | - |
A.13:8 - Rationale
The core answers which System qualifies for a local kind and holds its assignment. The profile answers separate questions about that holder’s capabilities. A Grade summarizes the profile for communication; it does not replace the detailed characteristic results needed by an assurance use.
The five characteristics are FPF engineering choices. A domain profile develops their operational definitions for its intended use; its measurement methods and any Grade mapping need their own stated basis. The following research contributions inform different agency questions:
- Levin’s scale-free cognition paper (2019) proposes an account of individuality and goal-directedness while allowing more than one useful agent definition.
- Levin’s TAME proposal (2022) describes a rich space of capacities rather than one privileged scalar of cognition. It motivates distinguishing capabilities instead of treating “agent” as a sufficient comparison.
- Friston et al.’s active-inference account of agency (2013) is a historical anchor concerning prior beliefs about action in an embodied account. That question differs from defining A.13’s five measurement scales or Grade mapping.
A.13:9 - Relations
- Builds on:
A.1 Holonic Foundation: Supplies the constructive and acting-eligibility conditions for recognizing the holder as aU.System.A.2 System-Role Kinds and Assignments: Distinguishes an exact local system-role kind, classification by that kind, and an obtainingU.SystemRoleAssignment.A.12 Acting-Side Externalization and Reflexive Split: Identify Who Acts on What: Separates acting and changed participants when that distinction matters; its precise account is conditional on the claims needed.
- Coordinates with:
B.2 Meta-Holon Transition (MHT): Changed supervision and a documented agency-characteristic threshold crossing can raise a whole-reidentification question about an established prior whole. B.2 first tests whether that whole’s identity rule accommodates the change and whether the whole meets the receiving use. Only an established transition warrants the MHT declaration.B.3 Trust & Assurance Calculus: A profile can supply evidence-backed characteristic results when a named reliability, safety or other assurance argument uses them. Retain each result’s bearer, scale and basis; the didactic Grade is not the assurance input.D.2 Recognize Ethical Concerns Across Levels and Scopes (Multilevel Ethics): For a multilevel ethical concern, name the local gain, possible loss, affected levels or scopes and next subject question. State responsibility or accountability separately under the rule defining that relation and its participants; use an agency-characteristic input only if that rule calls for it.