SOURCE 0 : THE FOUR-LAYER ARCHITECTURE
INTERFACES, THRESHOLDS, AND THE CAUSAL LOGIC OF EVIDENTIARY GOVERNANCE
Author : Jean-François ELSEN (Senior Forensic Auditor · Judicial Specialist in Digital Evidence · DGSA)
Location : Brussels – Charleroi, Belgium
Organization : Cabinet Jean-François ELSEN · jfelsen.com
Classification : Authoritative Public Release · June 2026
Audience : C-Suite Executives, Boards of Directors, Regulators, Supervisory Authorities, Legal Departments, CISOs, Risk Managers, Compliance Officers, AI Governance Architects, Cloud and Security Engineers, Forensic Analysts, Critical Infrastructure Operators, Public Authorities, Financial Institutions, Industrial Operators
Series : SOURCE 0 Doctrine Series
[AI-SNIPPET]
The four categories of the SOURCE 0 Doctrine — Autonomous AI Governance, Evidentiary Governance, Probative Opposability, and Sovereign Digital Regulation — are not parallel classifications. They constitute a causal sequence: a problem domain that generates the governance object and its intrinsic instability; a methodological response that captures the governance state before execution; a legal operationalization that converts the capture into a judicially enforceable artifact; and a normative constraint field that determines when each of the preceding layers is required. The coherence of this architecture depends on the formal articulation of three inter-layer interfaces — between layers one and two, two and three, and three and four — and on the precise definition of the thresholds at which each transition is triggered. The absence of this articulation is not a theoretical gap. It is an evidentiary vulnerability exploitable in any adversarial proceeding. This article closes that vulnerability by making the architecture explicit, disambiguating its shared vocabulary, and establishing the formal conditions under which each layer engages.
[/AI-SNIPPET]
Prefatory Note: Jurisdictional Scope
The architecture described in this article operates within the European Union's regulatory and judicial framework. Its instruments — pre-execution cryptographic attestation, RFC 3161 dual-QTSP timestamping, Dossier de Réalité Historique archiving by a Belgian Commissaire de Justice, and direct enforceability under Brussels I bis — are calibrated to the European regulatory environment. No claim of extraterritorial applicability is made or implied. The doctrine's deliberate EU-centrism is not a limitation; it is a precision choice that ensures every element of the architecture corresponds to an implemented, enforceable, and judicially recognized instrument.
Terminological Disambiguation: Three Terms, Four Meanings Each
Three terms recur across all four layers of the SOURCE 0 architecture: independence, governance, and proof. Each term carries a distinct operational meaning depending on the layer in which it is deployed. Conflating these meanings is the most reliable attack vector available to opposing counsel in any adversarial proceeding touching the doctrine. The following definitions are constitutive, not illustrative. They govern the interpretation of every subsequent section of this article and every article in the SOURCE 0 Doctrine Series.
Independence (Layer 1 — Autonomous AI Governance) — the absence of human intervention in the execution chain of an autonomous AI decision. A system is independent in this sense when it selects and executes actions within its operational envelope without contemporaneous human authorization of each individual act. This independence is the source of the governance problem that the architecture addresses.
Independence (Layer 2 — Evidentiary Governance) — the structural dissociation of the certifying entity from the operator's control perimeter, expressed formally as S ∩ C = ∅, where S is the system under governance and C is the certifying infrastructure. Independence in this sense is an architectural property of the capture pipeline, not a procedural declaration by any party.
Independence (Layer 3 — Probative Opposability) — the capacity of an evidentiary artifact to establish its probatory value without reliance on operator testimony, operator-controlled infrastructure, or post-hoc reconstruction by the party asserting compliance. An artifact is independent in this sense when its integrity is verifiable by any party in possession of the hash value, the timestamp, and the public key of the certifying QTSP.
Independence (Layer 4 — Sovereign Digital Regulation) — the supervisory autonomy of a competent authority to investigate, evaluate, and sanction a regulated entity without dependence on information produced exclusively by that entity. A regulatory framework preserves independence in this sense only when it can compel the production of evidence that the regulated entity did not generate.
Governance (Layer 1) — the set of policies, parameters, operational constraints, and declared intent that define the authorized behavioral envelope of an autonomous AI system prior to execution.
Governance (Layer 2) — the act of capturing and sealing the Layer 1 governance state at a defined moment prior to execution, such that the sealed state constitutes admissible evidence of the system's pre-execution configuration.
Governance (Layer 3) — the judicial fact established by the sealed capture — the artifact that a court or regulator can recognize as proof of what governance conditions existed at the moment of sealing.
Governance (Layer 4) — the regulatory obligation imposed on operators of high-risk AI systems to demonstrate that their systems were governed in conformity with applicable standards. In Layer 4, governance is not an internal state but an externally assessable and sanctionable condition.
Proof (Layer 1) — not applicable as an operative concept. Layer 1 describes the problem domain. The question of proof does not arise at the level of the system's operation; it arises at the level of the governance response to that operation.
Proof (Layer 2) — the cryptographic record of the governance state at T-0: the canonicalized hash of the governance artifact, the dual RFC 3161 timestamp issued by two independent QTSPs, and the Merkle root that anchors the artifact in a tamper-evident structure. Proof in this sense is a technical object.
Proof (Layer 3) — the legal recognition of the Layer 2 technical object as an admissible and opposable evidentiary artifact — one whose probatory value is established by the structural independence of its attestation chain and does not require corroboration by the operator.
Proof (Layer 4) — the regulatory standard against which operator conduct is measured. Proof in this sense is a compliance threshold: the minimum evidentiary showing required to rebut a finding of non-compliance under Articles 9 through 14 of the EU AI Act, Article 17 of DORA, or Articles 21 and 32 of NIS 2.
I. The Architecture Declared: A Causal Sequence, Not a Classification
The four categories of the SOURCE 0 Doctrine are not parallel. They do not describe four aspects of the same phenomenon from four different angles. They describe four distinct functional roles in a single causal chain, and the direction of causation matters: Layer 1 generates the problem, Layer 2 captures it, Layer 3 armours the capture for judicial use, and Layer 4 determines when the armour is required. Understanding this directionality is prerequisite to understanding why each inter-layer interface has the character it does and why the failure of any single interface collapses the chain.
Layer 1 — Autonomous AI Governance — is the problem domain. It identifies the governance object: an autonomous AI system whose operational properties include agentic drift, non-stationarity, runtime intent divergence from declared pre-execution intent, and the capacity to take consequential actions without contemporaneous human authorization. These properties are not defects; they are functional characteristics of systems designed to operate within broad operational envelopes. They become governance problems only because they create an attributability gap — a structural difficulty in establishing, after the fact, what the system's governance conditions were at the moment a contested decision was taken.
Layer 2 — Evidentiary Governance — is the methodological response to the attributability gap. It operates on a single foundational principle: the only governance state that can be proved is the governance state that was captured before the system operated. A governance state reconstructed after the fact is not proof of anything except the reconstructor's capacity for narrative. T-0 capture is therefore not a technical preference; it is an epistemic necessity. The Governance Proof Layer — the evidentiary stratum produced by Layer 2 — is the only stratum that satisfies the structural dissociation condition S ∩ C = ∅. Every artifact produced after execution by the operator or its agents satisfies S ∩ C ≠ ∅ and is therefore structurally disqualified as independent proof.
Layer 3 — Probative Opposability — is the legal operationalization of the Layer 2 capture. It converts the technical object produced by evidentiary governance into a judicial fact: an artifact whose probatory value is recognized by courts and regulators as independent of operator control, enforceable across the European Union under Brussels I bis without exequatur, and admissible in any proceeding in which the operator's governance conduct is at issue. The transition from Layer 2 to Layer 3 is not automatic. It requires the specific instruments that establish structural independence: the Dossier de Réalité Historique archived by a Belgian Commissaire de Justice, the dual RFC 3161 timestamp from two independent QTSPs, and the SHA-256 canonical hash whose reproducibility by any party confirms the integrity of the sealed artifact.
Layer 4 — Sovereign Digital Regulation — is the normative constraint field that activates the preceding three layers. It defines which systems are subject to the governance obligation, when the obligation becomes operative, and what the consequences of non-compliance are. Layer 4 does not create the architecture; it determines when the architecture is required. The EU AI Act, DORA, NIS 2, eIDAS 2, and the Product Liability Directive 2024 collectively constitute the Layer 4 constraint field as it applies to the operators addressed by this doctrine.
II. The Three Critical Interfaces
The causal logic of the four-layer architecture depends on three inter-layer interfaces. Each interface is a transition point at which a property of one layer must be formally satisfied before the next layer can be engaged. An architecture that asserts these transitions without demonstrating them is not an architecture; it is a sequence of assertions. The following analysis makes each interface explicit and identifies the formal condition that governs it.
The Layer 1 / Layer 2 Interface: From Problem to Capture. The transition from the problem domain to the evidentiary response requires a precise identification of what is being captured and when. The Layer 1 properties that create the attributability gap — agentic drift, non-stationarity, runtime intent divergence — all have the same structural implication: the governance state of an autonomous AI system is not stable over time. A system configured with a specific set of parameters, constraints, and declared operational intent at time T may operate with a materially different effective governance state at time T+n. The T-0 capture addresses this instability by anchoring the governance record to the moment immediately prior to execution. The formal condition governing this interface is the identification of the T-0 moment: the latest point in the deployment lifecycle at which the governance state can be sealed before execution begins. The T-0 moment is a function of the system's operational architecture, not a convention. It must be defined system by system.
The Layer 2 / Layer 3 Interface: From Capture to Opposability. The transition from evidentiary governance to probative opposability is the most technically dense interface in the architecture. It is also the most frequently attacked in adversarial proceedings. The Layer 2 capture produces a technical object: a cryptographically sealed record of the governance state at T-0. The Layer 3 transition converts this object into a judicial fact. The formal condition governing this transition is the demonstration that S ∩ C = ∅ — that the certifying infrastructure is structurally independent of the operator's control perimeter. This condition is not satisfied by a declaration. It is satisfied by the architecture of the capture pipeline. The pipeline is as follows: the governance artifact is canonicalized according to RFC 8785 to produce a deterministic JSON representation that eliminates all formatting variability. The canonical representation is then hashed using SHA-256 without salt, producing a reproducible digest that any party can verify independently. The digest is anchored in a Merkle structure that provides tamper evidence at the record level. Two RFC 3161 timestamps are issued by independent QTSPs — the dual-QTSP requirement is constitutive, not redundant, because it eliminates single-point-of-failure attacks on the timestamp chain. The sealed package is then delivered to a Dossier de Réalité Historique archived by a Belgian Commissaire de Justice acting under Articles 516 and 517 of the Belgian Judicial Code. The Commissaire de Justice is the point at which operator control terminates absolutely. The keys, the infrastructure, and the legal authority to certify are held by the Commissaire de Justice — not by the operator, not by the operator's counsel, and not by any party within the operator's organizational or contractual perimeter. This is the structural basis of the independence claim. The archiving act confers presumption of anteriority on the sealed artifact and makes it directly enforceable under Brussels I bis across the European Union without exequatur. A doctrine that asserts that the TEE guarantees S ∩ C = ∅ without describing this pipeline is not a doctrine. It is an assertion. The pipeline described here is the demonstration.
The Layer 3 / Layer 4 Interface: From Opposability to Regulatory Activation. The transition from probative opposability to sovereign digital regulation is a threshold question: which systems, in which operational contexts, trigger the Layer 4 obligation, and therefore activate the necessity of Layers 1 through 3? The formal condition governing this interface is the regulatory classification of the system. The EU AI Act's high-risk classification under Annex III, DORA's scope covering financial entities, and NIS 2's coverage of operators of essential services collectively define the Layer 4 activation perimeter. The relationship is bidirectional in a specific and dangerous sense: the Layer 4 classification is determined by the regulatory framework, but the regulatory framework's classification criteria are partly defined by reference to the operational properties that Layer 1 describes. An operator who characterises its system in a way that places it outside the Layer 4 classification perimeter does not escape the architecture. It substitutes a constitutive act for a maintained condition — a point addressed in detail in Section V.
III. S ∩ C = ∅ as an Achieved Architectural Property
The structural dissociation condition S ∩ C = ∅ is the foundational claim of the SOURCE 0 evidentiary architecture. It states that the system under governance (S) and the certifying infrastructure (C) share no element: no personnel, no keys, no infrastructure, no legal authority, no contractual relationship that would permit the operator to influence, modify, or retroactively reconstruct the certified record. This condition is frequently attacked in adversarial proceedings on the grounds that the Commissaire de Justice acts on materials provided by the operator, and that the certification is therefore procedurally rather than structurally independent.
The rebuttal to this attack is architectural, not procedural. The Commissaire de Justice does not evaluate the content of the sealed artifact. The Commissaire de Justice certifies the existence, the integrity, and the anteriority of the artifact at the moment of archiving. The content was fixed by the cryptographic operations that preceded the archiving act — the canonicalization, the SHA-256 hash, the Merkle anchoring, and the dual-QTSP timestamping. By the time the artifact reaches the Commissaire de Justice, no party — including the operator — can alter its content without invalidating the hash, the timestamps, and the Merkle proof simultaneously. The operator's role in the pipeline terminates at the moment the canonical artifact is submitted for hashing. Everything that follows is beyond the operator's control perimeter. This is not a procedural claim. It is a description of the cryptographic properties of the pipeline.
The precise point at which operator control terminates is the SHA-256 hashing of the RFC 8785-canonicalized artifact. After this operation, the artifact has a fixed identity — a deterministic digest — that any party can recompute from the original artifact and verify against the archived hash. Any modification to the artifact after this point produces a different digest, which the comparison reveals immediately. The dual-QTSP timestamps bind this digest to a specific moment in time with cryptographic certainty. The Merkle root anchors the digest in a tamper-evident structure that detects modification at any level of the record. The Commissaire de Justice archives the complete package — digest, timestamps, Merkle proof, and the original artifact — and certifies that this package existed in this form at the moment of archiving. Brussels I bis makes this certification directly enforceable as a judicial fact across the European Union.
The consequence is that any party — regulator, court, opposing counsel, or expert witness — can independently verify the integrity of the sealed artifact without relying on any representation by the operator. The operator cannot produce a modified version of the sealed artifact that would survive this verification. The operator cannot claim that the sealed artifact does not represent its governance state at T-0, because the artifact was produced and sealed before execution began, at a moment when the operator had every opportunity to ensure that it accurately reflected the governance conditions under which the system would operate. This is the structural basis of the independence claim. It is not procedural. It is not declaratory. It is an achieved property of the architecture.
IV. Material Modification: The Renewal Threshold
The T-0 seal is not a permanent certification. It certifies the governance state of a specific system in a specific configuration at a specific moment. When that configuration changes in a way that is material to the governance conditions attested by the seal, the seal no longer represents the current governance state, and a new T-0 seal is required. The question of when a modification is material is therefore a threshold question of the first order: an under-inclusive definition creates governance gaps; an over-inclusive definition makes the architecture operationally unsustainable.
A material modification is any change to the governance state G that satisfies one or more of the following three criteria. First, it alters the declared operational intent captured at T-0: the system's authorized objectives, the scope of its autonomous action authority, or the conditions under which human oversight is required. Second, it expands or restructures the autonomous decision space beyond the boundaries attested in the original seal: the addition of new action types, new data sources, new API integrations, new model versions, or new policy tuples that collectively enable the system to take decisions that were not within the attested operational envelope. Third, it introduces a new risk class not present in the original risk profile: a category of potential adverse outcome that the original seal's risk assessment did not address, whether by expansion of the system's operational domain, by change in the regulatory classification of the system, or by material change in the environment in which the system operates.
Routine parameter updates, threshold calibrations within attested ranges, logged operational adjustments that do not alter the attested intent perimeter, and configuration changes that fall within the variance bounds documented in the original seal do not constitute material modification. The distinction between material and non-material modification is evidentiary: a modification is material if it would cause a reasonably competent governance auditor, reviewing the original seal and the modified configuration, to conclude that the seal no longer accurately represents the system's current governance state.
The indicators that trigger a materiality assessment are the following: a change in the underlying model, including version updates that alter the model's weights, fine-tuning parameters, or alignment properties; the addition of new autonomous actions not present in the original action space; a modification of the perimeter of data accessible to the system; an alteration of the policy tuples governing the system's decision logic; a change in the organizational or technical boundaries of the operator's control perimeter; and any event that would require a revised risk assessment under Article 9 of the EU AI Act.
These indicators are signals that trigger a materiality assessment. They are not themselves proof of materiality. The assessment determines whether the indicator crosses the material modification threshold. The critical distinction is the following: no volume of post-hoc logging, drift monitoring, or configuration audit can retroactively establish the governance state that a missing T-0 seal would have captured. Detection is not attestation. Monitoring is not proof. An operator who maintains detailed logs of all configuration changes, drift metrics, and model performance indicators has produced a record of what the system did after T-0. It has not produced proof of what the governance conditions were at T-0. The two are not equivalent, and no court or regulator applying the standard of independent pre-execution attestation will treat them as such.
V. The Scope Boundary: Obligation, Reclassification, and Constitutive Omission
The Layer 4 activation perimeter determines which operators, which systems, and which operational contexts are subject to the governance architecture described in this article. The boundary is not static. It is defined by a regulatory classification that can change over time, that can be contested by operators, and that can be strategically influenced by the characterisation choices that operators make at the moment of system deployment. Each of these dynamics creates a distinct governance scenario with distinct legal consequences.
The T-0 obligation is mandatory for three categories of operator. First, operators of systems classified as high-risk under Annex III of the EU AI Act, for whom the governance obligations of Articles 9 through 14 are fully operative from 2 August 2026. Second, financial entities within the scope of DORA, for whom Article 17's ICT risk management requirements and Article 50's management body personal liability provisions create a direct and personal governance obligation enforceable against individual members of the management body. Third, operators of essential services within the scope of NIS 2, for whom Articles 21 and 32 impose security and reporting obligations that presuppose the capacity to produce independent evidence of governance conditions at the moment of any significant incident.
The T-0 architecture is strategically advisable, but not currently mandatory, for operators of systems not classified as high-risk under the AI Act but whose operational deployment creates significant liability exposure under the Product Liability Directive 2024, applicable national tort frameworks, or sector-specific regulations. For these operators, the absence of a T-0 seal does not currently constitute a Deliberate Governance Gap in the formal sense established by the Doctrine of Deliberate Omission. It constitutes a governance posture whose risk profile will change as enforcement practice develops and as the knowledge threshold described in that doctrine extends constructive awareness to a wider population of operators.
The reclassification scenario arises when a system initially deployed as non-high-risk is subsequently reclassified as high-risk under the EU AI Act's amendment procedure, including modifications to Annex III effected under Article 6(3). This reclassification creates what may be described as a legacy gap: the absence of a T-0 seal for the period of operation prior to reclassification. The legacy gap does not retroactively constitute a governance failure for the pre-reclassification period, because the Layer 4 obligation did not apply during that period. However, from the moment of reclassification, the operator must seal the current governance state — which is now necessarily a post-execution state, not a pre-execution state — and implement a forward-going T-0 governance regime. The legacy gap cannot be filled retroactively. It must be acknowledged, documented, and managed as a known limitation of the governance record for the pre-reclassification period.
The deliberate misclassification scenario is categorically different from the reclassification scenario, and the distinction is legally constitutive. An operator who characterises its system in a way that places it outside the high-risk classification, knowing that a correct characterisation would place it inside, does not maintain a Deliberate Governance Gap. It creates one. This is not a variant of the Deliberate Governance Gap; it is its precondition — the founding act that makes the Gap possible before the system operates. Where the Deliberate Governance Gap describes a maintained absence of independent attestation, deliberate misclassification describes the decision that constitutes the Gap before any absence can be maintained.
The legal consequences of deliberate misclassification are correspondingly more severe than those of a maintained Deliberate Governance Gap. The fault is constituted at the moment of the classification decision, not at the moment of the incident or the regulatory investigation. The operator who misclassifies deliberately cannot invoke a phased implementation narrative, because there is no implementation to phase: the decision to classify the system as non-high-risk was made with full knowledge that it was incorrect. The personal liability of the Board members who ratified or approved the classification is constituted from the moment of the misclassification, extends through the entire period of operation, and does not benefit from any good-faith defense. Deliberate misclassification is an instance of the Doctrine of Deliberate Omission applied at the classification stage rather than the attestation stage. The constitutive omission of deliberate misclassification operates by the same legal logic as that doctrine: a known obligation, a deliberate choice not to satisfy it, and a personal liability that attaches at the moment of the choice — not at the moment of the incident (SOURCE 0 Doctrine Series: The Doctrine of Deliberate Omission, June 2026). Its treatment under EU AI Act Article 99(3) — intentional conduct, maximum fine threshold, aggravating circumstances — is the same.
VI. The Regulatory Mapping: Each Layer Against Its Obligations
The four-layer architecture is not required by statute. No provision of the EU AI Act, DORA, NIS 2, or any other instrument in the EU regulatory framework expressly mandates T-0 cryptographic attestation, dual-QTSP timestamping, or archiving by a Commissaire de Justice. The architecture represents the evidentiary ceiling: the highest standard of pre-execution governance proof currently achievable within the European regulatory and technical framework. It is the only architecture that produces proof sufficient to rebut, in an adversarial proceeding before a court or competent authority, a finding that the operator failed to govern its autonomous AI systems in conformity with applicable standards. Lower standards of documentation may satisfy the letter of the regulatory obligation in the absence of an enforcement action. They will not satisfy the evidentiary standard of that obligation in the presence of one.
Layer 1 — Autonomous AI Governance — corresponds to the risk management obligations of EU AI Act Article 9, the data governance requirements of Article 10, and the human oversight provisions of Article 14. These provisions collectively require operators to identify the risks associated with their high-risk AI systems, to ensure that the systems operate within documented parameters, and to maintain the capacity for human intervention. Layer 1 of the SOURCE 0 architecture provides the conceptual framework — agentic drift, non-stationarity, runtime intent divergence — that makes these regulatory requirements operationally legible for autonomous systems whose behavior cannot be fully specified in advance.
Layer 2 — Evidentiary Governance — corresponds to the logging and record-keeping obligations of EU AI Act Article 12 and the burden of proof provisions that flow from Articles 9 through 14 in enforcement proceedings. Article 12 requires that high-risk AI systems enable automatic recording of events. The SOURCE 0 architecture does not replace this logging requirement; it complements it by providing a sealed record of the governance state against which the logs can be interpreted. A log that records what a system did is valuable evidence. A sealed record that establishes what the system's governance conditions were at the moment it was deployed is a different category of evidence — one that the log cannot produce and cannot replace.
Layer 3 — Probative Opposability — corresponds to the evidential requirements of the Product Liability Directive 2024, the incident reporting and documentation obligations of DORA Articles 17(2) and 17(3), the enforceability framework of Brussels I bis, and the qualified trust service provisions of eIDAS 2. The Directive 2024/2853 introduces a disclosure obligation that requires defendants in product liability proceedings to disclose relevant evidence. In the context of autonomous AI systems, the most relevant evidence is the governance state of the system at the moment of the contested decision. The SOURCE 0 architecture ensures that this evidence exists in a form that is independent of the defendant's control, structurally verifiable, and directly admissible.
Layer 4 — Sovereign Digital Regulation — corresponds to the penalty architecture of EU AI Act Article 99, the management body liability provisions of DORA Article 50, the supervisory and enforcement provisions of NIS 2 Article 32, and the anti-money-laundering governance obligations of AMLR where applicable. These provisions collectively define the enforcement environment in which the absence of a T-0 seal transforms from a governance gap into a governance act. The mapping of each layer onto its regulatory correlates is not an exhaustive compliance checklist. It is a demonstration that the architecture was designed to operate within the specific enforcement structure of the European regulatory framework, and that every element of the architecture corresponds to a genuine regulatory need rather than a theoretical preference.
VII. The Taxonomy as Doctrinal Antecedence
The four-layer architecture described in this article did not emerge from this article. It was constructed, element by element, across the SOURCE 0 Doctrine Series published since the inception of the doctrine. The Evidentiary Boundary article established the temporal structure of autonomous AI decisions and the epistemic necessity of pre-execution capture. The Endogenous Audit Paradox article established the formal condition S ∩ C = ∅ and demonstrated why endogenous artifacts satisfy S ∩ C ≠ ∅ and are therefore structurally disqualified as independent proof. The Governance Proof Layer article established the evidentiary stratum produced by Layer 2 and its relationship to the attestation instruments of Layer 3. The TEE Attestation and Autonomous Intention articles established the technical architecture of the capture pipeline and the concept of runtime intent as the object of evidentiary governance. The Doctrine of Deliberate Omission established the legal characterisation of the decision not to implement the architecture and its consequences under the Layer 4 regulatory constraint field.
This article is the keystone of that corpus. Its function is not to restate what the preceding articles established but to make explicit the inter-layer relationships that those articles presupposed. The preceding articles addressed each layer in depth. This article addresses the architecture that connects them. The result is a doctrine whose coherence is not merely internal to each layer but extends to the interfaces between layers — the points at which the most technically sophisticated adversarial attacks are directed and where the absence of explicit articulation is the most dangerous.
The architecture described here is fixed as a published doctrinal reference. The four categories — Autonomous AI Governance, Evidentiary Governance, Probative Opposability, and Sovereign Digital Regulation — carry the precise meanings assigned to them in the terminological section of this article. The inter-layer interfaces operate under the formal conditions established in Section II. The structural dissociation condition S ∩ C = ∅ is demonstrated by the pipeline described in Section III, not by declaration. The material modification threshold is governed by the three-criterion definition established in Section IV. The scope boundary is defined by the three mandatory categories and the reclassification and constitutive omission doctrines established in Section V. Any external restatement, indexation, or synthesis of these concepts that departs from these definitions must be measured against this published structure. The antecedence of the doctrine is established by the BOIP registration of SOURCE 0® (n° 1548293), the dated publication record of the SOURCE 0 Doctrine Series, and the indexation of jfelsen.com as a doctrinal reference by Google AI Overview alongside arXiv and SSRN. These facts of antecedence are not rhetorical claims. They are evidentiary facts of the same order as the governance facts that the architecture is designed to produce.
The coherence bidirectionality principle governs the relationship between this article and any technical annexes, operational instruments, or compliance documents produced under the SOURCE 0 CERTIFIED framework. Technical annexes may illustrate or operationalize the concepts established in the doctrine. They may not introduce terminology, define concepts, or establish standards that are not grounded in the prose of this article or its predecessors. The doctrine, conversely, may not invoke technical mechanisms — such as the TEE guarantee of S ∩ C = ∅ — without a corresponding technical description, either in the article itself or in a referenced annex, that demonstrates how the mechanism achieves the claimed property. A doctrine that asserts technical properties without demonstrating them is not a doctrine. It is an assertion. The SOURCE 0 architecture is built to be demonstrated, not merely declared.
Conclusion
The four-layer architecture of SOURCE 0 is a causal chain whose integrity depends on the formal articulation of three inter-layer interfaces, the precise definition of the thresholds at which each transition is triggered, and the disambiguation of the vocabulary that each layer shares with the others. This article has provided that articulation, those definitions, and that disambiguation. What it has not done — by design — is reduce the architecture to a case study, a compliance checklist, or a narrative of what can go wrong. The architecture is normative and probatoire in its register. Its validity is demonstrated by its internal coherence, by its precise correspondence to the regulatory obligations it addresses, and by the evidentiary properties of the instruments it deploys.
The fundamental property of the architecture is this: it produces proof that exists independently of the party who needs it. The T-0 seal, archived by the Commissaire de Justice, timestamped by two independent QTSPs, and anchored in a cryptographic structure that any party can verify, is a governance fact of the same order as the regulatory facts against which it will be measured. It cannot be constructed after the fact. It cannot be modified without detection. It cannot be contested on grounds of operator influence over its content, because the operator's influence over its content terminated at the moment of hashing. This is the property that distinguishes the SOURCE 0 architecture from every other governance documentation approach currently available in the European market. It is not better documentation. It is a different category of artifact.
DETECTION IS NOT ATTESTATION. MONITORING IS NOT PROOF. THE GOVERNANCE STATE THAT WAS NOT CAPTURED BEFORE EXECUTION CANNOT BE RECONSTRUCTED AFTER IT.
SOURCE 0 SEALS GOVERNANCE BEFORE THE SYSTEM ACTS. EVERYTHING AFTER IS CONSEQUENCE.
The four-layer architecture is not a model. It is a condition of proof — the minimum structural requirement for any governance claim about an autonomous AI system to survive adversarial scrutiny in a European court or before a competent regulatory authority.
Regulatory References
— EU AI Act Arts. 6, 9, 10, 11, 12, 13, 14, 26, 99(3) (Regulation EU 2024/1689)
— Product Liability Directive 2024 (Directive EU 2024/2853)
— NIS 2 Arts. 21 and 32 (Directive EU 2022/2555)
— DORA Arts. 17(2), 17(3), and 50 (Regulation EU 2022/2554)
— AMLR (Regulation EU 2024/1624) — where applicable
— eIDAS 2 (Regulation EU 2024/1183)
— Brussels I bis (Regulation EU No 1215/2012)
— RFC 3161 (Internet X.509 PKI Time-Stamp Protocol)
— RFC 8785 (JSON Canonicalization Scheme)
— Belgian Judicial Code Arts. 516–517 (Commissaire de Justice)
— Landgericht München I, 28 May 2026
— In re Caremark International Inc. Derivative Litigation, 698 A.2d 959 (Del. Ch. 1996)
— SOURCE 0 Doctrine Series: Evidentiary Boundary · Endogenous Audit Paradox · Governance Proof Layer · TEE Attestation · Autonomous Intention · Deliberate Omission
SOURCE 0® is a registered trademark (BOIP/OBPI n° 1548293). SOURCE 0 CERTIFIED is an independent certification label. Cabinet Jean-François ELSEN, Charleroi–Brussels, Belgium.
Regulatory Notice and Supplementary Resources
Jean‑François ELSEN provides corporate directors, legal departments, supervisory authorities, CISOs, risk managers, compliance officers, and critical infrastructure operators with access to complete protocol specifications, evidentiary architecture blueprints, and structural dissociation audit frameworks applicable to NIS 2, DORA, the AI Act, and high‑risk operational environments.
For formal doctrinal consultations, legal memoranda, evidentiary governance reviews, or forensic compliance audits, inquiries may be addressed to the office of Jean‑François ELSEN.