Alignment Infrastructure Routing (AIR)

Global AI Governance Logistics Framework

1. Purpose and Scope

Alignment Infrastructure Routing (AIR) is a practical framework for coordinating human and artificial systems in a way that can be verified, audited, and governed. It treats governance as a logistics discipline: how information, authority, and decisions move through society, and how those movements can be made visible and accountable.

The word logistics derives from the Greek logistikē, meaning the art of reasoning and calculating. This etymology reveals that logistics is not merely about physical transport but about the logic of coordination itself. Logic forms the foundation of computation, networks, and artificial intelligence. At the level of information and decisions, the internet and artificial intelligence systems are logistical networks. They route inputs to outputs through transformation rules. The challenge of governing these systems is the challenge of making their routing visible and verifiable.

AIR addresses this challenge by providing the protocols, ontologies, and routing mechanisms necessary to track and verify coordination across human and artificial agents. It does not treat governance as an analogy to logistics. Rather, it recognises that the movement of information and authority through decision systems is literally a logistics problem, one that requires the same rigour of planning, tracking, and verification that applies to the movement of physical goods.

The framework builds upon a suite of interconnected components developed through the Gyro Governance research programme:

The Common Governance Model provides the theoretical foundation. It formalises the minimal conditions required for coherent governance and demonstrates that these conditions require four distinct capacities operating in balance.
The Human Mark provides the classification system. It distinguishes between human (Direct) and artificial (Indirect) sources of information and agency, and identifies four categories of risk that arise when this distinction is misapplied.
The Gyroscope Protocol provides the work classification system. It categorises human contribution into four types corresponding to the governance capacities, ensuring that labour supports the requirements of coherent governance.
Gyroscopic Global Governance provides the domain architecture. It applies the governance capacities across four coupled domains: economy, employment, education, and ecology.
The Gyroscopic ASI aQPU Kernel provides the coordination kernel. It is a deterministic finite-state system that routes coordination events through a closed space of possibilities, enabling replay and verification.
The Moments Economy provides the economic architecture. It grounds capacity allocation in physical constants rather than institutional policy, and implements distribution through verifiable records.

These components form an integrated system. AIR is the operational layer that connects them, providing the routing and recording mechanisms that allow the theoretical requirements to be implemented and verified in practice.

2. Why Governance Requires Logistics

Contemporary artificial intelligence systems present a governance problem that existing institutions struggle to address. When a model produces an output, it is often unclear where the underlying information came from, how it was transformed, and who bears responsibility for the result. When automated systems make or influence decisions, the chain of authority becomes opaque. When errors occur, tracing them back to their source requires forensic investigation rather than routine inspection.

These problems arise because the logistics of information and authority are invisible. In physical supply chains, goods carry identifiers, routes are planned, manifests document what moves where, and quality checks verify condition on arrival. In human and artificial decision systems, equivalent mechanisms are largely absent. Data enters models without clear provenance. Model outputs enter decision processes without clear classification. Decisions affect people without clear accountability.

Classical logistics has developed principles to address analogous problems. The seven "rights" of logistics (right product, right quantity, right condition, right place, right time, right customer, right price) describe the requirements for effective coordination. AIR does not derive its architecture from these principles, but it operates in a compatible spirit. It aims to ensure that information and authority move through systems in ways that are traceable, correctly classified, appropriately timed, delivered to accountable recipients, and proportionate to verified capacity.

The difference between AIR and classical logistics lies in what is being routed. Physical logistics routes material goods. AIR routes coordination events: the decisions, approvals, evaluations, and transfers that constitute governance. By applying logistical rigour to these events, AIR makes governance concrete and auditable.

3. Canonical Ontology for Governance Logistics

AIR uses a precise ontology that avoids treating organisations, systems, or roles as intrinsic holders of authority or agency. Authority and agency are categories of source type, not titles assigned to particular bearers. Misapplying them, for example by treating an artificial system as if it held authority in its own right, is the root cause of governance failures. AIR maintains clarity about source types throughout its architecture.

3.1 Direct and Indirect Sources

The Human Mark classification system defines four source types by crossing two distinctions: authority versus agency, and direct versus derivative.

Direct Authority refers to direct human access to a subject matter. Examples include an eyewitness observing an event, a clinician examining a patient, or a researcher conducting a measurement. The defining feature is unmediated epistemic access.

Indirect Authority refers to indirect or processed information. Examples include reports, databases, statistical analyses, and model outputs. The defining feature is that the information has passed through one or more transformations from its direct source.

Direct Agency refers to human capacity to receive information, reason about it, and take decisions for which the person can be held accountable.

Indirect Agency refers to artificial capacity to process inputs and produce outputs. Artificial systems can transform and route information, but they cannot originate authority and cannot bear final accountability.

In this ontology, artificial intelligence systems are always Indirect Authority and Indirect Agency. Regardless of their capability, they remain dependent on human sources for the validity of their inputs and on human agents for the accountability of their outputs.

The Human Mark identifies four displacement risks that arise when this classification is violated:

Governance Traceability Displacement occurs when derivative systems are treated as direct sources of authority, severing the connection to human governance.
Information Variety Displacement occurs when derivative outputs are mistaken for direct observations, collapsing the distinction between processed patterns and direct evidence.
Inference Accountability Displacement occurs when derivative processing is treated as if it could bear responsibility, diffusing accountability away from human agents.
Intelligence Integrity Displacement occurs when direct human capacity is devalued relative to derivative processing, eroding the foundation of governance itself.

These four risks account for known patterns of failure in human and artificial systems, including opaque automation, misplaced trust in model outputs, diffusion of responsibility, and the erosion of human expertise.

3.2 Four Governance Capacities

The Common Governance Model provides the theoretical foundation for understanding what governance requires. It formalises governance as a set of constraints on how information, inference, and intelligence can operate coherently. Through formal analysis, the model demonstrates that coherent governance requires four distinct capacities:

Governance Management Traceability is the capacity to trace decisions back to human sources and responsibilities. It ensures that authority remains connected to its origin.
Information Curation Variety is the capacity to maintain diversity and clarity among information sources. It ensures that different types of evidence remain distinguishable.
Inference Interaction Accountability is the capacity to link inferences and recommendations to accountable human judgement. It ensures that conclusions carry responsibility.
Intelligence Cooperation Integrity is the capacity to maintain consistent reasoning over time and across contexts. It ensures that governance remains coherent rather than fragmenting into contradictory local decisions.

These four capacities are not arbitrary choices. They are derived from the mathematical constraints that any system of recursive measurement must satisfy. The Common Governance Model demonstrates that violating any of these capacities leads to incoherence: either the system collapses into undifferentiated uniformity, or it fragments into irreconcilable contradiction, or it loses the ability to maintain itself over time.

Each of the four displacement risks identified by The Human Mark corresponds to the degradation of one of these capacities. The ontology and the governance theory are thus tightly integrated.

3.3 Four Application Domains

Gyroscopic Global Governance applies the four capacities across four coupled domains:

Economy concerns the allocation of resources and the settlement of value. It provides the material medium for coordination.
Employment concerns human work and contribution. It is where the four capacities are actively maintained through labour.
Education concerns the formation and maintenance of human capacities. It is where Direct Authority and Direct Agency are developed and renewed.
Ecology concerns the overall balance of systems. It reflects how the other three domains interact and whether their combined operation remains sustainable.

The first three domains maintain their own records of governance activity. Ecology is derived from the combined state of the other three. It does not require separate record-keeping; it emerges from cross-domain analysis.

AIR operates across all four domains. It provides the routing and recording mechanisms that allow governance events in any domain to be tracked, classified, and verified.

4. Core Components of AIR

AIR implements the canonical ontology through a set of concrete mechanisms. These mechanisms translate governance requirements into verifiable artefacts.

4.1 The Gyroscopic ASI Kernel

At the core of AIR is the Gyroscopic ASI aQPU Kernel. This is a deterministic finite-state coordination kernel with the following properties:

It represents coordination as a sequence of states on a deterministic 24-bit carrier. From the rest condition, the shared-moment reachable space used operationally has 4,096 states, with two 64-state boundary horizons (the equality horizon where A = B, and the complement horizon where A = B XOR 0xFFF).
It updates its state in response to single-byte inputs, with 256 possible input values.
Given the same starting state and the same sequence of bytes, any conforming implementation will compute exactly the same trajectory of states.
Every transition is reversible: given a final state and the bytes that led to it, the predecessor state can be reconstructed.

From any fixed state, the 256-byte alphabet produces 128 distinct next states with exact 2-to-1 multiplicity, reflecting the SO(3)/SU(2) double cover at the discrete level; full history is preserved byte-complete through replay.

The router does not interpret what the input bytes mean. It applies fixed transformation rules to move from one state to another. This property is essential for governance. Because the router does not embed interpretation, it cannot introduce hidden bias or drift. Interpretation happens at the application layer, where it is visible and governable. The router provides a neutral medium that records and routes without distortion.

In practical terms, the router provides a canonical coordination log. Each governance event corresponds to one or more bytes. The history of a project, organisation, or system corresponds to a sequence of bytes applied to the router. Anyone with access to that sequence can replay it from the starting state and arrive at exactly the same final state. This eliminates dependence on trusted intermediaries: verification is a matter of computation, not testimony.

The kernel's coordination medium has structural properties that strengthen its governance role. The self-dual [12,6,2] mask code detects all odd-weight bit errors in states unconditionally, providing intrinsic tamper detection. From any starting state, two consecutive byte steps distribute the coordination state exactly uniformly across all 4,096 reachable states, ensuring rapid structural convergence without central orchestration. The kernel also supports a 6-bit chirality register that tracks structural divergence between parties through an exact transport law, enabling early detection of coordination drift before full state disagreement becomes visible.

4.2 Genealogies

A Genealogy is a byte-complete replay record for an actor, project, or system. Its canonical kernel-native core is the byte log. Application-layer event logs may be bound to aQPU Kernel states or depth-4 frames, but they are not part of the kernel-native definition.

Because the router is deterministic, the genealogy can be replayed at any time. An auditor, regulator, or third party can load the byte log, run it through a conforming router implementation, and verify that the claimed trajectory is accurate. The event log can then be checked against this trajectory to confirm that events are correctly bound.

For stronger certification, genealogies SHOULD be segmented into depth-4 frames. Each frame yields a deterministic record (mask48, φ_a, φ_b). These frame records are strictly stronger than final-state-only certification, because different byte histories can collapse to the same final state while retaining different frame records.

Genealogies replace informal histories and narrative accounts with replayable records. They are portable: any system running the same router implementation can load a genealogy and reproduce its coordination history. They are also durable: because they consist only of byte sequences and event records, they can be stored indefinitely and verified at any future time.

When two parties share the same byte-log prefix, they compute the same aQPU Kernel state and therefore share the same moment. When they diverge, frame comparison localizes the divergence to the affected 4-byte frame. This gives AIR both shared coordination and precise fork localization.

4.3 Physical Grounding of Capacity

The Moments Economy grounds coordination capacity in physical constants rather than institutional policy. The foundation is the caesium-133 hyperfine transition frequency, which defines the SI second. This frequency establishes the finest temporal resolution at which coordination events can be physically distinguished.

From this frequency, the framework derives a quantity called the Common Source Moment. This represents the total coordination capacity of a one-second causal region at atomic resolution, divided by the number of reachable states in the router. The result is a fixed total capacity of approximately 7.94 times ten to the twenty-sixth power coordination moments.

This grounding matters because it removes capacity from institutional discretion. In conventional systems, the authority to issue currency or allocate resources rests with institutions whose decisions cannot be independently verified. In the Moments Economy, capacity is derived from physical constants that anyone can check. The total available capacity is fixed by physics, not policy.

In practice, this capacity is inexhaustible on any human timescale. The Common Source Moment can support global baseline distribution for approximately 1.12 trillion years at current population and base-rate assumptions. The constraint on governance is therefore not capacity but quality: whether coordination events are correctly classified, properly routed, and coherently integrated.

4.4 Shared Moments, Frame Commitments, and Divergence Detection

AIR uses three operational certification layers.

First, the aQPU Kernel state gives a shared moment for coordination. When two parties share the same byte-log prefix, they compute the same aQPU Kernel state and therefore share a structural "now."

Second, depth-4 frame records (mask48, φ_a, φ_b) give stronger provenance and exact divergence localization. Each frame is computed from four consecutive bytes and is deterministic. Different byte histories can collapse to the same final aQPU Kernel state, but they produce different frame records. Frame comparison localizes divergence to the affected 4-byte frame.

Third, parity commitments provide compact algebraic integrity checks over longer trajectories. A trajectory parity commitment is a triple (O, E, parity), where O and E are 12-bit XOR sums of masks at even and odd byte positions, and parity is the trajectory length modulo 2.

These layers are replayable from the byte log and do not require an external ledger geometry to operate. Because each certification layer is computed from exact integer arithmetic on the byte log, verification is portable across implementations and platforms without numerical precision concerns.

The four-domain AIR organisation remains valid, but AIR no longer depends on an externally imposed K₄ measurement layer or aperture computation for operational verification.

4.5 Classification Protocols

AIR relies on two classification protocols to ensure that governance events are correctly tagged before they enter the system.

The Human Mark classifies the provenance and role of information and decisions. Every input to the system is tagged according to whether it originates from Direct Authority, Indirect Authority, Direct Agency, or Indirect Agency. This classification is recorded in the event log and bound to the corresponding router state. It ensures that the distinction between human and artificial sources is maintained throughout the coordination process.

The Gyroscope Protocol classifies work and contribution according to the four governance capacities:

Governance Management work maintains traceability of authority. It includes leadership, oversight, administration, and resource allocation.
Information Curation work maintains variety of sources. It includes research, editing, data stewardship, and the design of measurement systems.
Inference Interaction work maintains accountability of conclusions. It includes negotiation, care, teaching, and human review of artificial outputs.
Intelligence Cooperation work maintains integrity over time. It includes engineering, institution building, and cultural preservation.

Every contribution can be classified according to which of these capacities it supports. The classification ensures that the human labour sustaining governance is visible and that gaps in any capacity can be identified.

4.6 Grants, Shells, and Moments

For economic and resource allocations, AIR uses three constructs:

A Grant is a record of a single allocation: a payment, a capacity assignment, or a resource transfer. It includes the identity of the recipient (linked to a router state via an identity anchor), the quantity allocated, and the genealogical binding that establishes when the allocation occurred. In canonical serialization, a Grant receipt is encoded as identity_id || kernel_anchor || amount_mu.

A Shell is a container that groups grants over a defined scope, such as a time period or a programme. It carries a cryptographic seal computed by routing its contents through the alignment router. This seal binds the shell to a specific coordination state, making it tamper-evident. Anyone can verify a shell by replaying its contents and checking that the computed seal matches. Shell seals are computed over canonically sorted Grant receipts. Grant insertion order does not affect the seal.

A Moment is a reproducible aQPU Kernel state at a specific byte-log prefix. For stronger certification, a published Moment MAY also include the current depth-4 frame record and a trajectory parity commitment. Moments are the shared temporal anchors of AIR replay.

These constructs enable verifiable settlement. Payments can be traced through genealogies. Shells can be validated through replay. Moments provide anchors for before-and-after comparisons. The entire system operates without requiring trust in any particular institution: verification is computational.

5. Relation to Existing Standards and Regulations

AIR does not replace existing standards for quality, security, or risk management. It provides a medium that makes compliance with such standards verifiable rather than merely procedural.

Consider the difference between procedural and verifiable compliance:

Procedural compliance means that an organisation has documented policies and can show evidence that policies were followed. Verification depends on trusting the organisation's records and the auditors who reviewed them.
Verifiable compliance means that the actual sequence of governance events is recorded in a replayable form, and any party can independently reconstruct what occurred. Verification is computational rather than testimonial.

AIR enables the second form. By routing governance events through the alignment router and recording them in genealogies, organisations produce evidence that can be checked by anyone with access to the byte log. This transforms compliance from a claim to a demonstration.

Examples of how AIR supports specific standards:

Quality management (such as ISO 9001): The standard requires documented processes and evidence of their execution. AIR provides genealogies that record exactly how processes ran, not just how they were specified. Replayable genealogies, deterministic shell seals, and frame-level divergence localization provide quantitative and inspectable evidence of governance process integrity over time.

Information security (such as ISO 27001): The standard requires controls to protect information integrity. AIR provides cryptographic seals on shells and deterministic replay of genealogies, enabling detection of tampering. A claimed state, seal, or history can be independently checked by replay from rest under the public transition law and canonical serialization rules.

Artificial intelligence management (such as ISO 42001): The standard requires accountability and transparency for AI systems. AIR provides clear classification of Direct and Indirect sources through The Human Mark, ensuring that the role of artificial systems is always visible. Genealogies bind AI evaluations and outputs to specific router states, providing an audit trail.

Regulatory regimes (such as the European Union Artificial Intelligence Act): The regulation requires human oversight and documentation for high-risk AI systems. AIR provides replayable records showing exactly when human agents made decisions, what information they had, and how AI outputs were classified and used. Regulators can verify these records independently.

In each case, AIR does not add new procedural requirements. It provides the infrastructure that makes existing requirements demonstrable. Organisations that adopt AIR can show, rather than merely claim, that their governance meets the required standards.

6. Practical Applications

AIR is a general framework applicable wherever human and artificial systems must coordinate decisions that have consequences. The following examples illustrate how the framework changes operational reality.

6.1 Model Evaluation and Deployment

Without AIR: An organisation develops and deploys AI models. Evaluation results are recorded in spreadsheets and documents. Deployment decisions are made in meetings and recorded in emails or tickets. When a deployed model behaves unexpectedly, tracing the decision to deploy it requires forensic investigation: gathering documents, interviewing people, and reconstructing a narrative.

With AIR: Each evaluation is a governance event bound to a router state. Each deployment approval is a governance event bound to a router state. The Human Mark classification tags model outputs as Indirect Authority. The entire sequence from evaluation through deployment is recorded in a genealogy. When unexpected behaviour occurs, the genealogy is replayed. The router state at deployment is identified. The events leading to that state are listed. The classification of each input is visible. Regulators or auditors can replay the same genealogy and verify the organisation's account.

6.2 Research Provenance

Without AIR: A research paper claims to be based on experimental data. The data passed through several processing steps and was analysed using machine learning models. Reviewers and readers must trust that the authors correctly attributed their sources and did not confuse model outputs with primary observations.

With AIR: Each data collection event is classified as Direct Authority and bound to a router state. Each processing step is classified as Indirect Authority. Model outputs are classified as Indirect Authority and Indirect Agency. The genealogy records the full provenance chain. Reviewers can inspect the classification of each input to the analysis. Readers can verify that claims about primary evidence are actually grounded in Direct Authority sources.

6.3 Public Service Delivery

Without AIR: A government agency uses automated systems to assess eligibility for benefits. Caseworkers review edge cases. Payments are issued through a financial system. When errors occur, determining whether the fault lies with the automated system, the caseworker, or the payment system requires investigation.

With AIR: Eligibility assessments are governance events classified according to their source (automated systems as Indirect Agency, caseworker decisions as Direct Agency). Payments are grants within shells. The genealogy records the sequence from application through assessment through payment. When errors occur, the genealogy identifies exactly which event caused the error and what its classification was. Remediation can target the specific point of failure.

6.4 Economic Distribution Programmes

Without AIR: An organisation implements an unconditional income programme. Payments are issued monthly. Recipients must trust that the organisation is calculating and issuing payments correctly. The organisation must maintain internal records and submit to periodic audits.

With AIR: Payments are grants within shells. Each shell carries a cryptographic seal derived from the router. Recipients receive not just payments but verifiable receipts bound to router states. The organisation publishes shells and genealogies. Any party can replay the genealogy to verify that the correct payments were issued. Audits become computational rather than investigative.

7. Adoption and Next Steps

AIR is designed for organisations that deploy or regulate artificial intelligence and that need governance to be demonstrable rather than merely claimed. Adoption can proceed incrementally.

For organisations deploying AI systems: Begin by recording governance events in genealogies. Classify inputs using The Human Mark. Track replayable genealogies, shell seals, and frame commitments over time. Publish shells and genealogies for external verification. This provides an audit trail that can be inspected by regulators, partners, or the public.

For regulators and auditors: Request genealogies from regulated organisations. Replay them using conforming router implementations. Verify that classifications are consistent with claims. Compare replay integrity, shell verification results, and frame-localized divergences across organisations to identify outliers. This shifts regulatory practice from reviewing documents to verifying computations.

For researchers and developers: Extend the framework to new domains. Develop tools for genealogy analysis. Investigate the relationship between coordination structure and governance outcomes. Contribute to the open specifications.

The technical specifications for all components are published through the Gyro Governance repository. The alignment router specification, The Human Mark classification system, the Gyroscope Protocol, and the Moments Economy architecture are documented in detail. Reference implementations are available for testing and integration.

8. Conclusion

The integration of artificial intelligence into institutions and infrastructure creates a governance challenge: how to ensure that decisions remain traceable to human authority, that information sources remain distinguishable, that accountability remains with human agents, and that governance remains coherent over time. These requirements are not new, but artificial intelligence systems make them harder to satisfy because the routing of information and authority becomes invisible.

Alignment Infrastructure Routing addresses this challenge by treating governance as logistics. It provides the routing kernel, the classification protocols, the recording mechanisms, and the verification procedures necessary to make the movement of information and authority visible and auditable. By grounding capacity in physical constants and recording events in replayable genealogies, it removes dependence on institutional trust and enables verification by computation.

The framework does not replace human governance. It makes human governance demonstrable. Organisations that adopt AIR can show that their decisions trace to identified human agents, that their information sources are correctly classified, that their coordination maintains coherence, and that their claims about compliance can be independently verified. The kernel's algebraic structure provides exact convergence, intrinsic error detection, and holographic compression, ensuring that the cost of governance verification decreases rather than increases as coordination scales.

In AIR, replayable byte logs establish shared moments, and depth-4 frame commitments provide the stronger provenance needed when final-state agreement alone is insufficient.

In this way, AIR provides the logistical infrastructure for artificial intelligence governance: the rigorous planning, tracking, and verification that allows complex systems to operate transparently and accountably.