DLP Specification
Part III: World Model & Governance Theory

§8 Control-Theoretic Foundation

Formal control theory grounding. Conant-Ashby → Ashby → Beer VSM → LeCun. Predictive Composition Pattern.

§8 Control-Theoretic Foundation

v2.0.0 · Locked · L1 · March 19, 2026

Purpose

The DLP primitive architecture converges with four established results from seven decades of control theory, cybernetics, and machine intelligence — all pointing toward organizational world model convergence. This section traces the derivation chain that grounds the protocol in these results, maps the structural correspondence to Beer's Viable System Model, operationalizes Ashby's requisite variety as nine irreducible governance dimensions, and formalizes the state prediction isomorphism between organizational governance and world model research, showing how the organizational world model must satisfy all four requirements simultaneously.

The argument is deductive, not analogical. Each step narrows the architectural requirements for any governance infrastructure that operates as an organizational world model. The DLP contribution is convergent implementation: the first governance infrastructure that satisfies all four requirements simultaneously.

Foundation

The Derivation Chain

Four established results define what an organizational world model must be. Each builds on the previous, narrowing the design space from general principle to specific architecture.

StepPrincipleSourceRequirementDLP Response
1Conant-Ashby TheoremR.C. Conant & W.R. Ashby (1970)Every good regulator of a system must be a model of that system. Governance infrastructure must BE a model of the organization — not a set of rules applied from outside.The substrate is an organizational model. Nineteen primitives across five tiers represent organizational state; state transformations capture organizational dynamics.
2Law of Requisite VarietyW.R. Ashby (1956)V(R) ≥ V(D): the model's variety must match the system's complexity. A governance framework with fewer independent dimensions than the organizational reality it governs will fail to represent some decision contexts.Nine Tier 1 primitives are the specific, testable requisite variety for organizational governance — nine independent governance dimensions validated by removal, merger, and sufficiency tests (§4.2).
3Viable System ModelS. Beer (1972–1985)Five subsystems (S1–S5) are necessary and sufficient for organizational viability, operating recursively at every scale. Missing subsystems produce predictable pathologies.Behavioral invariants (B1–B10) formalize the S5 policy function (§5). Deviation measurement implements S3/S3*. Substrate profiles (EAS, BAS, PAS) instantiate the same primitive structure at every organizational scale.
4World Models as Predict-Imagine-Act EnginesY. LeCun (2022; 2026)A world model predicts S(t+1) from S(t) and action A(t), enabling planning without exhaustive trial. The model must project forward, not just record.The state transformation model — State(t) → [Trigger + Action] → State(t+1) — is structurally identical to the world model prediction formalism. Predictive Composition (§8.7) formalizes forward projection over primitives.

Step 1 → Step 2. If governance must be a model of the system (Conant-Ashby), the model must have sufficient variety to represent the system's states (Ashby). A model with fewer degrees of freedom than the system it models loses information about some system states. A governance system that cannot represent a state cannot govern it.

Step 2 → Step 3. If the governance model requires requisite variety (Ashby), there exist specific structural components constituting the minimum viable model for organizational systems (Beer). Beer derived these from neurophysiology: S1 (operations), S2 (coordination), S3 (control), S3* (audit), S4 (intelligence), S5 (policy). The claim is that these five functions are necessary — not merely useful — for viability, and that their absence produces specific, diagnosable pathologies.

Step 3 → Step 4. If the governance model requires specific structural subsystems (Beer), and those subsystems map to a formal state-transformation architecture (LeCun), then the governance model operates as a world model — predicting organizational state transitions, detecting deviations between predicted and actual states, and enabling corrective action. The substrate IS the organizational world model.

VSM Structural Correspondence

Beer's five systems are not analogies for DLP components. They are structural correspondences: both frameworks solve the same control problem — maintaining viability in complex environments — and arrive at functionally equivalent subsystems from independent starting points.

VSM SystemFunctionDLP MappingCorrespondence Type
S1 — OperationsAutonomous primary activities; the units that do the workWork + Capacity + Evidence primitives; operational state transformationsStructural
S2 — CoordinationDampening oscillations between operational units; scheduling, conflict resolutionConstraint primitive; coordination patterns between substrate instancesStructural
S3 — ControlResource allocation, optimization, accountability for operational performanceAuthority + Account + Commitment primitives; deviation measurementStructural
S3 — Audit*Sporadic direct investigation of operational reality; bypasses regular reportingEvidence primitive (verification function); continuous audit infrastructureExtension
S4 — IntelligenceEnvironmental scanning, future modeling, adaptation planningIntent + Orientation primitives; Predictive Composition Pattern (§8.7)Structural + Extension
S5 — PolicyIdentity, purpose, ultimate authority; balances S3 (inside/now) vs. S4 (outside/future)Behavioral invariants (B1–B10); Intent primitive (organizational purpose)Structural
Algedonic ChannelEmergency signals bypassing hierarchy when viability is threatenedDeviation alerting against behavioral invariants; Signal Capture mechanism (B7, B8)Extension

Original contribution. No prior published work formally connects Beer's System 4 to modern world model research. Beer described S4 as the organizational intelligence function — environmental scanning and future modeling — but never formalized its internal mechanism. The correspondence between S4 and the world model prediction formalism is identified here: S4 IS the organizational world model function. §8.7 formalizes what S4 does internally as the Predictive Composition Pattern.

The three Extension rows mark where DLP extends Beer's framework beyond its original formulation. S3* moves from sporadic to continuous verification — a technology constraint of 1972 that no longer holds. S4 gains a formal internal mechanism through Predictive Composition. The algedonic channel moves from binary pain/pleasure to quantified deviation measurement with thresholds and authority-routed escalation.

The S3/S4 Homeostat

Beer's central structural insight: organizational viability depends on the dynamic balance between S3 (operational stability, "here and now") and S4 (environmental intelligence, "there and then"). S5 exists to govern this balance.

ConditionDiagnosisDLP ManifestationDetection
S3 overwhelms S4Operationally efficient but strategically blindAuthority + Account dominate; Intent + Orientation inactiveHigh operational conformance, low adaptation signals
S4 overwhelms S3Perpetually adaptive but operationally chaoticIntent shifts frequently; Commitment instability; Work completion dropsLow operational conformance, high strategy variance
S3/S4 balancedViableAll nine Tier 1 primitives active; graduation progressing; deviations within boundsBehavioral invariants satisfied; evidence accumulating

Requisite Variety and the Nine Primitives

Ashby's Law states that the variety of the regulator must equal or exceed the variety of the disturbances it controls: V(R) ≥ V(D). Beer applied this qualitatively to organizations: governance must match organizational complexity. Neither Ashby nor Beer specified what constitutes the requisite variety for organizational governance — what the independent dimensions are, how many are needed, or how to verify sufficiency.

For seventy years, requisite variety has been a powerful design principle with no operational answer for governance. The nine Tier 1 primitives are the specific, testable answer: nine independent governance dimensions, validated by three attack vectors.

Removal test. Each primitive is individually necessary. Removing any single primitive leaves at least one governance question unanswerable (§4.2).

Merger test. No viable mergers exist. Each primitive occupies an orthogonal semantic position — no primitive can be expressed as a function of the remaining eight (§4.2).

Sufficiency test. The nine collectively express any organizational decision state. No governance scenario requires a tenth independent dimension (§4.2).

PrimitiveGovernance DimensionVariety Contribution
IntentPurpose and directionDistinguishes purposeful action from purposeless activity
CommitmentResponsibility and obligationDistinguishes binding acceptance from aspiration
CapacityResource realityDistinguishes feasible from infeasible
WorkTransformationDistinguishes executed effort from planned effort
EvidenceEpistemic groundingDistinguishes substantiated from unsubstantiated claims
DecisionSelection among optionsDistinguishes chosen path from available alternatives
AuthorityLegitimacy of actionDistinguishes authorized from unauthorized action
AccountGovernance stateDistinguishes current state from historical state
ConstraintUniversal rulesDistinguishes permitted from prohibited action

Each row represents an independent regulatory dimension. Removing any row reduces the governance model's variety below the organizational reality it must regulate — a direct violation of Ashby's Law.

Graduation as Variety Engineering

Beer's practical mechanism for managing the variety gap between organization and environment is variety engineering: attenuation (filtering variety upward) and amplification (multiplying regulatory signals downward). The graduation path is a variety engineering mechanism.

Graduation StageVariety OperationMechanism
Option A: Human routesMaximum attenuation for AI, maximum amplification for humanSubstrate captures and structures (attenuates raw variety); human makes all routing decisions (amplifies through judgment)
Option B: Substrate proposes, human confirmsBalanced variety managementSubstrate reduces the decision space to ranked proposals (attenuation); human confirmation amplifies the selected option into action
Option C: Substrate routes within delegationMaximum amplification for substrate, attenuation through policy boundsSubstrate amplifies organizational capacity through autonomous routing; behavioral invariants attenuate by constraining the action space
Stages 0–4Progressive variety transferEvidence accumulation justifies transferring variety management from human to substrate — each stage requires demonstrated competence at the previous level

Convergent Evolution

The DLP primitive architecture was not derived from cybernetics, compliance frameworks, or machine learning theory. It was derived from governance practice — grants administration, compliance work, and organizational decision-making. The structural correspondence with these traditions is convergent: independent frameworks solving the same problem arrive at functionally equivalent structures from independent starting points.

COSO → VSM Structural Correspondence. Original contribution: No published work maps COSO's Internal Control Framework to Beer's Viable System Model. This correspondence is identified here for the first time.

COSO Component (1992/2013)VSM System (1972–1985)Structural Correspondence
Control EnvironmentS5 — PolicyBoth define organizational identity, ethical values, and governance structure
Risk AssessmentS4 — IntelligenceBoth scan for threats and model future scenarios
Control ActivitiesS3 — ControlBoth implement policies through operational mechanisms
Information & CommunicationS2 — CoordinationBoth ensure appropriate information flows between organizational units
Monitoring ActivitiesS3* — AuditBoth perform verification that operational reality matches expectations

Two traditions for organizational governance — compliance (COSO, arising from accounting fraud scandals) and cybernetics (VSM, arising from neurophysiology and control theory) — converge on the same five structural requirements. The substrate implements both: behavioral invariants formalize S5/Control Environment; deviation measurement implements S3*/Monitoring Activities; Authority + Constraint implement S3/Control Activities.

Four-Tradition Convergence. Original contribution: No prior publication connects these four independent traditions.

TraditionStarting PointStructural DiscoveryKey ResultPeriod
CyberneticsControl theory, feedback, informationFive necessary systems for organizational viability; requisite variety; recursive structureBeer's VSM: specific structural requirements for viability1956–1985
Machine Learning TheoryOptimization theory, representation learningTraining objectives systematically break invariances that real intelligence must preserveChlon (2026): invariances must be given, not learned (§9)2022–2026
Audit & Governance PracticeProfessional practice, regulatory complianceNine irreducible dimensions of organizational decision context; state transformation captureDLP: governance infrastructure with specific primitive architecture2024–2026
Compliance FrameworksInternal control, risk managementFive components of effective control; three lines of defenseCOSO (1992/2013): structural requirements for organizational control1992–2013

None cites the others. Each discovers the same requirement from its own domain: organizations need architecturally imposed governance structure with sufficient variety to match their complexity, operating recursively at every scale.

The closest historical precedent for cybernetic governance infrastructure applied to organizational coordination is Project Cybersyn (Chile, 1971–1973) — Beer's own attempt to implement VSM as national economic governance. Cybersyn demonstrated that the governance infrastructure Beer theorized was implementable in principle and that the technology was the binding constraint. The constraint no longer exists.

Governance

This section is owned by Cam (founder, GrytLabs). Changes to the derivation chain, VSM correspondence, or Predictive Composition pattern require explicit decision with patent-impact assessment. Cross-references to §5 (behavioral invariants) and §9 (conservation laws) require synchronized updates across all three sections.

Substance

§8.5 State Prediction Isomorphism

The organizational world model and the AI world model share the same formal structure. This is not analogy — it is structural identity across different state representations.

World Model ComponentAI FormulationOrganizational FormulationDLP Mechanism
StateLearned latent representation S(t)Full primitive tuple — all instantiated primitives across Tiers 1–5, depending on deployment profileTyped record containing all active primitive instances at time t
ActionAgent action A(t)Organizational decision or external triggerTrigger + Action with authority context and constraint evaluation
Transition functionF(S(t), A(t)) → S(t+1)Organizational dynamics producing the next governance stateT(State, Trigger, Action) → State(t+1) with complete lineage
Prediction errorDeviation between predicted and observed next stateDeviation between projected and actual organizational stateDeviation measurement against Intent and Orientation as target states
Evidence / ObservationSensory input updating the modelOrganizational signals, reports, audit findingsEvidence primitive with truth type classification (§6)

The structural difference is the state representation. AI world models operate in learned latent space — compressed representations that are powerful but opaque. The DLP substrate operates in explicit primitive space — nineteen typed dimensions across five tiers that preserve organizational invariances because they are architecturally given, not learned from data. §9 establishes mathematically why the explicit representation is necessary: training objectives under log loss systematically break the invariances that governance must preserve.

The organizational world model observes under the Open World Assumption: the organization may contain states that the model has not yet captured. What is not recorded may still be true. The enforcement layer (§5) operates under the Closed World Assumption through the shapes graph: what is not explicitly permitted is prohibited. The world model layer and the enforcement layer maintain complementary epistemic stances — observation is open, validation is closed.

§8.6 DLP Extensions Beyond VSM

The structural correspondence does not mean the substrate is a VSM implementation. DLP extends Beer's framework in nine specific ways.

VSM LimitationDLP ExtensionMechanism
Diagnostic framework onlyOperational governance infrastructureBeer teaches diagnosis of what is missing. The substrate captures what is happening, making diagnosis continuous rather than episodic.
Variety as qualitative principleNine primitives as specific, testable requisite varietyBeer and Ashby established variety as a design principle. DLP operationalizes it as countable, enforceable governance dimensions.
Recursive model (diagnostic)Recursive implementation (substrate profiles)Beer's recursion identifies the same five systems at each level. Substrate profiles (EAS, BAS, PAS) instantiate the same primitive structure at every organizational scale, enforcing consistency.
S4 as functional descriptionPredictive Composition Pattern (§8.7)Beer described S4's function (environmental intelligence) but not its internal mechanism. DLP formalizes S4 as forward state projection over primitives.
S3* as sporadic auditContinuous verification via Evidence primitiveBeer accepted sporadic monitoring as a technology constraint of 1972. The constraint no longer exists.
Binary algedonic channelQuantified deviation measurement with thresholdsBeer's pain/pleasure binary becomes specified deviation vectors, threshold values, and authority-routed escalation (B7, B8).
No formal state transformation modelState(t) → [Trigger + Action] → State(t+1) with full lineageBeer described organizational state conceptually. DLP captures state transformations as typed, auditable records with complete primitive context.
No truth type systemAuthoritative → Declared → Derived epistemic infrastructureBeer's framework has no mechanism for distinguishing human assertion from AI interpretation. The truth type system (§6) prevents AI outputs from silently becoming organizational truth.
No AI integration modelOption A/B/C with graduation stagesBeer's framework predates operational AI. The substrate provides governed progressive autonomy within delegation bounds (§12).

§8.7 Predictive Composition: S4 Internal Mechanism

Beer described System 4's function — environmental scanning and future modeling — but never specified its internal mechanism. What representations does S4 use? How does it generate predictions? How does it measure consistency between projected and actual states?

The Predictive Composition Pattern answers these questions. It is a composition pattern over existing primitives that runs the state transformation model forward: given the current state (full primitive tuple across all instantiated tiers), project deviation against Intent and Orientation as target states. The question it answers: given current trajectory, what deviations is the organization heading toward?

This makes budgets, estimates, and forecasts substrate-native — not bolted-on analytics but forward projections over the same primitive infrastructure that captures current state.

Predict-Imagine-Act Correspondence. The Predictive Composition Pattern maps to LeCun's three-phase world model loop.

PhaseAI World ModelOrganizational World ModelDLP Mechanism
PredictForward projection from current state given actionProject primitive tuple forward under candidate decisionpredict(state, action) → projected_state over full primitive tuple
ImagineEvaluate alternative action trajectoriesCompare projected states against Intent and Orientation as targetsDeviation measurement between projected states and target states
ActCommit to trajectory with lowest expected costCommit to path with lowest expected governance deviationDecision primitive with projected evidence basis, authorized by Authority

SDK Constraints.

  • MUST: The SDK exposes a predict(state, action) → projected_state operation over the full primitive tuple.
  • MUST: Projected states carry the Derived truth type (§6). They are model outputs, not authoritative records.
  • MUST NOT: Predictive projections create binding commitments without human decision. A projection is an input to Decision, not a substitute for it.
  • DESIGN SPACE: Projection horizon, deviation thresholds, and trigger conditions are profile-configurable. Different deployment profiles (EAS, BAS, PAS) configure these parameters for their organizational context.

Predictive Composition is an architectural pattern, not a locked specification. This subsection establishes the pattern's conceptual foundation and SDK constraint boundaries. Full algorithmic specification — projection methods, deviation aggregation, confidence propagation — is deferred to implementation. An SDK team implements the constraint boundaries defined here; the projection internals are implementation choices within those boundaries.

Boundaries

This section specifies the derivation chain, structural correspondences, and the Predictive Composition pattern. It does NOT specify: the primitives themselves (§4), the behavioral invariants (§5), the conservation laws (§9), the substrate implementation architecture (§26), or specific projection algorithms.

The state prediction isomorphism describes the formal relationship between AI and organizational world models. It is structural, not prescriptive — it identifies the correspondence without fixing implementation choices for projection methods or deviation measurement.

Positions

Locked. Four-result derivation chain: Conant-Ashby → Requisite Variety → VSM → World Models. Nine primitives as operational requisite variety. VSM structural correspondence with nine extensions. State prediction isomorphism between AI and organizational world models. S3/S4 homeostat as organizational viability balance. Predictive Composition Pattern as S4 internal mechanism with SDK constraints.

Original contributions. VSM → DLP structural correspondence (first published connection). COSO → VSM correspondence (first published). Four-tradition convergence (cybernetics, ML theory, governance practice, compliance frameworks). S4 formalization as Predictive Composition. DLP extensions beyond VSM (nine specific mechanisms).

Lineage

v1.0.0 (February 25, 2026): Base lock. Derivation chain, VSM correspondence, requisite variety operationalization, convergent evolution, state prediction isomorphism, DLP extensions.

v2.0.0 (March 19, 2026): Post-lock composition. Predictive Composition Pattern (§8.7) formalized with SDK constraints. S3/S4 homeostat table expanded. Four-tradition convergence table completed.

Commitments

SDK implementations MUST expose a predict(state, action) → projected_state operation over the full primitive tuple. SDK implementations MUST classify projected states as Derived truth type. SDK implementations MUST support profile-configurable projection horizons, deviation thresholds, and trigger conditions.

Coverage

All four derivation steps fully specified with source citations. VSM structural correspondence complete with nine extension mechanisms. State prediction isomorphism specified across five components. S3/S4 balance characterized across operational, adaptive, and viable conditions. Predictive Composition Pattern specified with three-phase correspondence and SDK constraints. Nine DLP extensions beyond VSM documented.

Addressing

Document ID: s08-control-theoretic-foundation. Part: III (World Model). Subsections addressable as §8.1–§8.7. Tables addressable by content (e.g., "Table 8.2.1: VSM → DLP Structural Correspondence"). Cross-references use [s08-control-theoretic-foundation.{section-id}] notation.

Attribution

Primary author: Cam (founder, GrytLabs). Research grounding and composition support: Claude (Anthropic). Historical research: Cybernetics literature (Ashby, Beer, Conant), machine learning theory (LeCun, Bengio, Yang, Chlon), COSO framework, Project Cybersyn.

Situational Frame

The derivation chain was developed through systematic research across 17 sprints, cross-referencing control theory (Ashby's law), cybernetics (Beer's VSM), and governance practice. The VSM correspondence emerged from pattern recognition across independent frameworks. The Predictive Composition Pattern was formalized in response to OI-22 (S4 formalization) and locked at v2.0.0. Four-tradition convergence composed from research synthesis and original contribution documentation.

Scope Governance

Core namespace: dlp. All derivation concepts and primitives reside in core namespace. Predictive Composition is an architectural pattern; implementation profiles (EAS, BAS, PAS) configure pattern parameters within the core namespace, not through separate implementations.

Framing

This section is the theoretical foundation of the DLP. It establishes why the primitive architecture is not arbitrary but convergent with seven decades of control theory, cybernetics, and AI research. The Predictive Composition Pattern operationalizes what Beer theorized but never implemented: the internal mechanism of System 4. The section anchors architectural decisions in first-principles reasoning from independent research traditions.

Adaptation

The Predictive Composition Pattern (§8.7) was added post-lock to answer a gap: Beer's VSM describes what S4 does (environmental intelligence) but not how. The pattern formalizes the mechanism by operationalizing LeCun's world model formalism at the organizational level. The S3/S4 homeostat table was expanded to provide diagnostic criteria for viability. Four-tradition convergence was composed when independent traditions (cybernetics, ML theory, governance practice, compliance) were recognized to solve the same structural problem.

Readiness

This section is ready for SDK implementation. The Predictive Composition Pattern defines three SDK constraints; the state prediction isomorphism is testable against concrete projections. Implementation teams should start with the constraint boundaries (MUST/MUST NOT) and treat projection internals as implementation choices within those boundaries. Dependencies: §4 (primitives), §5 (behavioral invariants), §6 (truth type system), §9 (conservation laws), §12 (governance activation).

Meaning Resolution

No meaning resolution for this document.

Perception Surface

No perception surface for this document. Control-theoretic concepts are abstract organizational principles; they do not interface with external systems directly. However, the state prediction isomorphism (§8.5) describes how the organizational world model observes organizational state — this observation flows through the Evidence primitive (§4) and Environment Interface (§4.6.3).

Temporal Governance

The Predictive Composition Pattern operates at organizational timescales — projecting from current state across decision horizons (days to quarters depending on profile configuration). Projection horizons, deviation measurement intervals, and trigger evaluation cycles are profile-configurable (DESIGN SPACE) but not locked at this level of specification. §13 (Policy Projection Surfaces) provides concrete time-bound manifestations.