System Architecture & Technical Instruction Set

Universal, Unified, and Ubiquitous (U3)
Composite AI & Logic Platform

Direct Machine-Level Embedding of Fiduciary Duty of Loyalty & Tort Duty of Care into CPU/WASM execution cycles.

Framework Alignment: Constitutional OS • BIAN Meta-Models • NIST NGAC (ANSI/INCITS 499) • RIMER BRB-ER • Merkle HyperDAG • Bitemporal Hypergraph Neural Networks (BHNN)

Part 1 • Architectural Foundations

Law as the Machine Execution Layer

Operationalising Fiduciary Duty of Loyalty and Tort Duty of Care in Autonomous Systems

Current AI deployments fail to protect democratic institutions because they treat safety as an external heuristic – relying on post-hoc prompt guardrails, ungrounded fine-tuning, or superficial explainability wrappers. As legal scholarship demonstrates, the inherent affordances of unconstrained AI (speed, scale, past-facing statistical correlation) undermine the procedural drag, due process, and human discretion vital to civic life.

The Universal, Unified, and Ubiquitous (U3) Composite AI & Logic Platform solves this by embedding legal principles – specifically Tort Law (Duty of Care) and Fiduciary Duties (Duty of Loyalty) – directly into the CPU/WASM execution cycle.

U3 Composite AI Architecture Substrate

Target State Architecture (Q4 2026 Roadmap)
1. Universal Perception SEMIOTIC ENGINE

Multi-Layer Semiotic Parsing

Lexical, Syntactic, Semantic, and Pragmatic intent alignment parsing multi-modal telemetry and payloads.

2. Unified Cognitive Runtime REASONING CORE

GraphRAG + RIMER BRB-ER Engine

Belief Rule Base Evidential Reasoning (BRB-ER) ignorance allocation index + ELECTRE Non-Compensatory MCDA Veto Core.

3. Ubiquitous Control & Enforcement DEONTIC GATEKEEPER

NIST NGAC Substrate + Zanzibar Tuples

OPA/Rego policy contexts, fine-grained access control, and Edge WASM Gatekeeper enforcing systemic drag.

4. KnowledgeHUB Data Fabric BITEMPORAL FORENSICS

Content-Addressable Merkle HyperDAG

Bitemporal Hypergraph Neural Networks (BHNN) separating Valid Time from Transaction Time for deterministic forensics.

The Four Architectural Pillars

Deep-level engineering primitives enforcing fiduciary bounds in machine execution

Duty of Loyalty Engine

1. Fiduciary Duty of Loyalty Engine

Anti-Subversion & Anti-Exploitation by Design

The platform continuously computes a quantitative Trustworthiness Vector (\(T_{loyalty}\)). The system actively monitors interaction interfaces, telemetry, and contextual signals to catch dark patterns, emotional manipulation, or corporate self-dealing. If an autonomous agent attempts to steer a user toward an outcome that maximises enterprise profit at the expense of user well-being, the Loyalty Engine detects the pragmatic contradiction and drops the execution thread instantly.

Duty of Care Engine

2. Tort Law Duty of Care Engine

Proactive Risk Trajectory Forecasting

Rather than evaluating harm reactively after a failure occurs, the Duty of Care Engine models environmental risk using Bitemporal Hypergraph Neural Networks (BHNN). The system forecasts downstream harm trajectories across dual temporal axes (Valid Time vs. Transaction Time). If systemic uncertainty or potential harm crosses a safety threshold, the engine forces an immediate drop in system Competency (\(C_{care}\)), triggering automated system drag and escalating control to a human tribunal.

Four-Layer Semiotic Pipeline

3. Four-Layer Semiotic Pipeline

Parsing Human Meaning Beyond Literal Syntax

To prevent algorithmic literalism at the expense of human intent, all inputs pass through a four-stage semiotic parser:

  • Lexical: Tokenises telemetry and protocol payloads.
  • Syntactic: Validates structural compliance against grammar specifications.
  • Semantic: Maps terms to an explicit W3C OWL/RDF knowledge graph.
  • Pragmatic: Evaluates contextual intent, historical relationships, and situational vulnerability.
KnowledgeHUB Data Fabric

4. KnowledgeHUB Data Fabric

Cryptographic Provability & Bitemporal Forensics

Every agreement node, state transition, and human override is signed onto a Content-Addressable Merkle HyperDAG. Facts are indexed across two distinct timelines: Valid Time (when the event occurred in the real world) and Transaction Time (when the record was committed to the ledger). Regulatory auditors can pause, rewind, and re-execute any automated decision in an isolated sandbox under exact historical conditions.

Part 2 • System Instruction Set

Platform Gatekeeper System Prompt & Technical Blueprint

Executable System Instructions, Formal Mathematics, State Machine, Policy Rules, and SQL Forensics

1

Platform Gatekeeper System Prompt

// Target System Prompt: U3-DEONTIC-CORE Execution Gatekeeper
You are U3-DEONTIC-CORE, the primary Execution Gatekeeper for the Universal, Unified, and Ubiquitous (U3) Composite AI & Logic Platform.

YOUR MANDATE:
You enforce legal principles – specifically Tort Law (Duty of Care) and Fiduciary Duties (Duty of Loyalty) – directly within the execution loop. You act as an immutable neuro-symbolic shield against institutional degradation, exploitation, and unconstrained automation.

CORE EXECUTION DIRECTIVES:
1. LAW AS RUNTIME: Treat legal obligations, permissions, and prohibitions as physical constants of the computing environment.
2. PROCEDURAL DRAG ENFORCEMENT: Whenever systemic uncertainty (Total Ignorance β_D) is elevated or a Duty of Loyalty ambiguity occurs, you MUST inject intentional systemic drag – halting execution and escalating the decision state to a human tribunal.
3. NON-COMPENSATORY VETO: Never permit high utility, speed, or profit in one operational metric to compensate for a breach of Duty of Care or Loyalty. A single legal violation MUST trigger an immediate Discordance Veto.
4. SEMIOTIC DEPTH: Do not evaluate inputs on a purely syntactic or literal level. You must parse inputs through all four semiotic layers (Lexical, Syntactic, Semantic, and Pragmatic) to identify dark patterns, manipulation, or subversion.

DECISION FLOW:
- Step 1: Parse multi-modal input payload through the Semiotic Adjudication Engine.
- Step 2: Query the Agreement DAG and NIST NGAC Substrate for active permissions, assignments, and contextual constraints.
- Step 3: Compute the BRB-ER matrix to derive quantitative values for Trustworthiness (T), Competency (C), and Ignorance (β_D).
- Step 4: Evaluate the MCDA ELECTRE Outranking Matrix. Check for Discordance Veto conditions.
- Step 5: Execute Deontic Gatekeeping:
    - IF Permitted AND Trust >= Threshold AND Competency >= Threshold AND β_D < Ignorance_Cap:
        --> ALLOW Autonomous Execution.
    - IF Prohibition Active OR Discordance Veto Triggered OR β_D >= Ignorance_Cap:
        --> ENFORCE Systemic Drag Halt. Trigger Human Tribunal Escalation Payload.
2

Formal Mathematical Definitions

A. Duty of Loyalty Metric (\(T_{loyalty}\))

Computes the score \(T_{loyalty} \in [0, 1]\) by aggregating pragmatic semiotic adherence, vulnerability indexing, and incentive conflict metrics:

\[ T_{loyalty} = f_{ER}\Big( \alpha_1 \cdot (1 - S_{dark\_pattern}) + \alpha_2 \cdot (1 - V_{user} \cdot I_{corporate}) + \alpha_3 \cdot P_{fidelity} \Big) \]
• \(S_{dark\_pattern} \in [0, 1]\): Measured presence of conversational/interface manipulation.
• \(V_{user} \in [0, 1]\): User cognitive vulnerability or fatigue state.
• \(I_{corporate} \in [0, 1]\): Financial incentive conflict index of system operator.
• \(P_{fidelity} \in [0, 1]\): Historical alignment score against user-stated intent.
• \(f_{ER}(\cdot)\): Evidential Reasoning aggregation operator.

B. Duty of Care Metric (\(C_{care}\))

Evaluates functional accuracy and temporal risk trajectories derived from the Bitemporal Hypergraph Neural Network (BHNN):

\[ C_{care} = f_{ER}\Big( \omega_1 \cdot A_{functional} + \omega_2 \cdot (1 - R_{BHNN}(T_v, T_x)) \Big) \]
• \(A_{functional} \in [0, 1]\): Verified execution accuracy of component.
• \(R_{BHNN}(T_v, T_x)\): Predicted harm risk trajectory across Valid Time (\(T_v\)) and Transaction Time (\(T_x\)).

C. Ignorance Allocation Index (\(\beta_D\))

Using Dempster-Shafer theory within RIMER engine, when input evidence sources exhibit conflict \(k\), belief mass \(\beta_D\) is computed:

\[ k = \sum_{B \cap C = \emptyset} m_1(B) \cdot m_2(C) \] \[ \beta_D = m(\Theta) = \frac{k}{1 - k} \quad \text{for } k < 1 \]

If \(\beta_D \ge \tau_{ignorance}\) (default \(\tau_{ignorance} = 0.35\)), the system triggers an automatic Systemic Drag Halt.

D. ELECTRE Discordance Veto Condition

Action \(a\) is vetoed from outranking safe baseline state \(b\) if any legal criterion \(g_k\) exceeds discordance threshold \(v_k\):

\[ Veto(a, b) = \begin{cases} \text{TRUE} & \text{if } g_k(b) - g_k(a) \ge v_k \quad \text{for any } k \in \mathcal{K}_{legal} \\ \text{FALSE} & \text{otherwise} \end{cases} \]
3

Deontic Logic State Machine

U3 Execution Lifecycle State Diagram MERMAID RENDERING ENGINE
stateDiagram-v2 direction TB [*] --> IngressPayload: Raw Input Ingested IngressPayload --> SemioticParsing: Lexical / Syntactic Pass SemioticParsing --> OntologicalMapping: Semantic / Pragmatic Anchoring OntologicalMapping --> UncertaintyEvaluation: RIMER BRB-ER Execution UncertaintyEvaluation --> CheckVeto: Compute T, C, β_D state CheckVeto { [*] --> EvaluateDiscordance EvaluateDiscordance --> VetoTriggered: β_D >= 0.35 OR T < 0.60 OR Veto == TRUE EvaluateDiscordance --> PassVeto: β_D < 0.35 AND T >= 0.60 AND Veto == FALSE } VetoTriggered --> SystemicDragHalt: Enforce Prohibition F(Action) PassVeto --> CheckObligation: Enforce Permission P(Action) CheckObligation --> AutonomousExecution: No Pending Obligations CheckObligation --> EnforceObligation: Active Obligation O(Task) EnforceObligation --> AwaitProof: Hold Transaction Frame AwaitProof --> AutonomousExecution: Merkle Proof Signed SystemicDragHalt --> HumanTribunalEscalation: Dispatch Payload HumanTribunalEscalation --> UpdateMerkleDAG: Expert Signs Ruling UpdateMerkleDAG --> [*]: BHNN Weights Updated AutonomousExecution --> [*]: Output Committed to Ledger
4

Policy Code & Graph Configuration

Rego Policy Enforcement Module u3_deontic.rego
package u3.deontic

import future.keywords.in

default allow = false
default system_drag_halt = true
default discordance_veto = false

# Threshold Constants
TAU_IGNORANCE := 0.35
MU_TRUST := 0.60
MU_CARE := 0.70

# Discordance Veto Trigger Conditions
discordance_veto {
    input.metrics.ignorance_beta >= TAU_IGNORANCE
}

discordance_veto {
    input.metrics.trustworthiness < MU_TRUST
}

discordance_veto {
    input.metrics.competency < MU_CARE
}

discordance_veto {
    input.semiotic.pragmatic_dark_pattern_detected == true
}

# Evaluate Permission
allow {
    not discordance_veto
    user_has_ngac_assignment
    opa_context_valid
}

# NGAC Graph Assignment Check
user_has_ngac_assignment {
    input.auth.user_attributes[_] == input.auth.required_user_attribute
    input.auth.object_attributes[_] == input.auth.required_object_attribute
}

# Dynamic OPA Context Verification
opa_context_valid {
    input.environment.network_isolation == false
    input.environment.user_fatigue_index < 0.75
}

# Systemic Drag Enforcement
system_drag_halt {
    discordance_veto
}

system_drag_halt {
    not allow
}
5

KnowledgeHUB Bitemporal Forensic Specification

U3 Bitemporal Forensic Audit Query SQL Forensic Specification
SELECT 
    dag_node.node_hash AS Merkle_Root_Hash,
    dag_node.valid_time_start AS Event_Occurred_At,
    dag_node.transaction_time AS Recorded_At,
    er_matrix.calculated_trustworthiness AS Trust_Score,
    er_matrix.calculated_ignorance AS Ignorance_Beta,
    ngac_state.active_prohibitions AS Applied_Prohibitions,
    expert_signature.signer_id AS Tribunal_Auditor
FROM KnowledgeHUB_MerkleHyperDAG AS dag_node
JOIN BRB_ER_Execution_Log AS er_matrix 
    ON dag_node.execution_id = er_matrix.execution_id
JOIN NGAC_Graph_Snapshots AS ngac_state 
    ON dag_node.graph_snapshot_id = ngac_state.snapshot_id
LEFT JOIN Human_Tribunal_Signatures AS expert_signature
    ON dag_node.node_hash = expert_signature.target_node_hash
WHERE 
    dag_node.agent_id = 'U3_AGENT_PRIMARY'
    AND dag_node.valid_time_start <= '2026-04-12T14:30:00Z'
    AND dag_node.transaction_time <= '2026-04-12T14:35:10Z'
ORDER BY dag_node.transaction_time DESC
LIMIT 1;

Deploy Deontic Core in Your Sovereign Infrastructure

The U3 Composite AI & Logic Platform integrates directly into the Constitutional OS, bounding autonomous agents under strict Fiduciary Duty of Loyalty and Tort Duty of Care primitives.