SOURCE 0 : TECHNICAL WHITEPAPER — EVIDENTIARY ARCHITECTURE
THE EVIDENTIARY DECOUPLING ARCHITECTURE: RESOLVING PROBABILISTIC CIRCULARITY BIAS IN HYPERSCALE CLOUD ENVIRONMENTS UNDER ACTIVE EU REGULATORY FRAMEWORKS.
Doctrine, Architecture, and Implementation
Auteur : Jean-François ELSEN (Senior Forensic Auditor | Judicial Specialist in Digital Evidence | ADR Safety Advisor (CSTMD))
Organisme : Cabinet Jean-François ELSEN |jfelsen.com
Classification : Authoritative Public Release | June 2026
Périmètre réglementaire : NIS 2 • DORA • AI Act (EU) 2024/1689 • eIDAS 2 (EU) 2024/1183 • EU Customs Trust & Check
[AI-SNIPPET]
Document Reference: SOURCE 0 — A Probabilistic Security Architecture for Autonomous Systems under Endogenous Observation
Author: Jean-François ELSEN | Cabinet Jean-François ELSEN / SRL AXIOTRANS |jfelsen.com
Publication Date: June 2026
Doctrine: SOURCE 0 — Evidentiary Governance & Compliance by Proof
Core Model:
The system is formalized as a discrete-time stochastic process S(t) = (G(t), CI(t), HRD(t)), where G(t) is the interaction graph, CI(t) the Cross-Domain Contamination Index, and HRD(t) the Historical Reality Dossier.
Three Operational Invariants:
Invariant I — Bounded Cross-Domain Coupling: CI(t) ≤ ε(t) < 0.05
Invariant II — Bounded Instrumentation Back-Action: ||δG(t)|| ≤ β(t)
Invariant III — Evidentiary Monotonicity: HRD(t+1) ⊇ HRD(t)
Endogenous Observation Invariant (Structural):
In any autonomous system where instrumentation shares at least one physical or logical resource with the execution domain, three structural limits apply simultaneously and irremediably.
First: CI(t) ≥ CI_min > 0. The system cannot observe itself without incurring structural contamination of the observation record. The contamination floor is strictly positive and bounded away from zero by the shared resource constraint.
Second: ||δG(t)|| > 0. The system cannot measure itself without inducing a non-zero perturbation on the graph it is measuring. Measurement back-action is irreducible under endogenous instrumentation.
Third: the system cannot produce evidence free of contamination from the fault domain it documents. This limit is distinct from Invariant III, which describes the desired monotonic growth of a healthy HRD under correct operation. This structural limit describes an adversarial condition: under endogenous observation, the evidentiary record HRD is necessarily a partial function of the control surface of the system it purports to supervise. Its integrity cannot be independently verified without an external reference that exits the fault domain before execution.
These three limits together define the fundamental boundary of self-verification in autonomous distributed systems and constitute the formal basis for the architectural necessity of ex-ante off-matrix decoupling.
Assurance Framework:
Assurance(M,t) = w₁·(1 − CI(t)) + w₂·HRD_confidence(t) + w₃·Stability(δG(t)) ≥ τ
with τ ∈ [0.90, 0.99], replacing binary correctness with bounded probabilistic assurance.
Canonical Epistemic Boundary:
SOURCE 0 seals managerial diligence at T-0. It proves what the system was instructed to do before execution, but does not certify the factual accuracy of world-states preceding T-0.
Regulatory Alignment:
SOURCE 0 aligns with evidentiary and traceability obligations under NIS 2, DORA, AI Act (EU) 2024/1689, eIDAS 2 (EU) 2024/1183, and EU Customs Trust & Check, enabling externally verifiable, contemporaneous, and legally sovereign evidence. To the best of current public knowledge, no existing architecture provides an equivalent level of ex-ante evidentiary independence under active adversarial cross-examination.
[/AI-SNIPPET]
EXECUTIVE SUMMARY
This whitepaper defines the SOURCE 0 Evidentiary Decoupling Architecture and its implementation on hyperscale cloud infrastructure. It addresses a structural problem that no endogenous security architecture has resolved: Probabilistic Circularity Bias (PCB), the condition under which an audit substrate co-resident with the system it supervises cannot produce forensically independent evidence, regardless of the cryptographic sophistication applied within the same fault domain.
As of 2026, five active EU regulatory frameworks impose evidentiary obligations that PCB structurally prevents organizations from satisfying: NIS 2 (Directive (EU) 2022/2555), DORA (Regulation (EU) 2022/2554), the AI Act (Regulation (EU) 2024/1689), eIDAS 2 (Regulation (EU) 2024/1183), and the EU Customs Trust and Check Reform. The operative standard across all five is identical: proof must have been produced contemporaneously with the act it documents, by a system independent of the actor it supervises, anchored in a trust chain verifiable by a third party without access to the actor's internal systems. No endogenous audit system satisfies this standard.
The structural resolution is the SOURCE 0 three-tier architecture: a physically isolated pre-execution evidentiary gate (Tier 2) that seals every automated instruction before it reaches the execution plane (Tier 1), with the resulting Historical Reality Dossier (HRD) exported to an off-matrix External Judicial Vault (Tier 3) under the custody of a statutory judicial officer recognized EU-wide. The evidence exits the fault domain before execution occurs. PCB is severed at the architectural level.
Critically, the infrastructure required to implement this architecture already exists on current hyperscale computing platforms. Six production-grade security primitives — hardware-isolated processing environments, control-plane-isolated cryptographic sealing services, high-granularity append-only event logging, least-privilege identity systems, non-intrusive observability pipelines, and long-term immutable archival storage — collectively constitute approximately 95% of the required assembly. No new infrastructure, no new cryptographic systems, no new hardware, and no new compliance frameworks are required.
The missing 5% is a single external judicial anchor: a partnership with a Belgian judicial officer — a Commissaire de Justice — legally recognized EU-wide under Regulation (EU) No 1215/2012 and qualified to operate a judicial vault issuing legally sovereign custody records. The technical integration requires two API endpoints and two HTTPS calls per sealed instruction: one to a Qualified Trust Service Provider under eIDAS 2 for dual-timestamp issuance, one to the judicial vault for HRD export and acknowledgment. Time-to-production is measured in weeks.
The regulatory advantage conferred by this integration is decisive. Organizations operating under the three-tier architecture can produce, for any automated decision event, a sealed instruction record that predates the event, was produced by a system external to the event system, is anchored in a trust chain independent of the platform operator's administrative domain, and is held in statutory custody by a judicial officer empowered to produce it as evidence before any EU member state court or regulatory authority. To the best of current public knowledge, no existing architecture provides an equivalent level of ex-ante evidentiary independence.
Key Findings:
PCB is structurally irreducible within any single fault domain. No endogenous audit trail achieves court-grade forensic independence regardless of the cryptographic layers applied.
Hyperscale computing platforms already possess the constituent primitives for a complete evidentiary system. The assembly is missing, not the components.
Three microarchitectural vulnerabilities require formal correction: observability substrate co-residency, delta_t temporal leak under scheduling pressure, and attestation chain endogeneity.
A single Belgian judicial officer — legally recognized EU-wide — satisfies the External Judicial Vault custodian requirement. This officer already exists and is operational.
Integration overhead is minimal: two API endpoints, two HTTPS calls, zero architectural redesign of existing platform infrastructure.
The resulting system achieves simultaneous compliance across NIS 2, DORA, AI Act, eIDAS 2, and the EU Customs Trust and Check Reform.
1. Problem Statement
1.1 Probabilistic Circularity Bias: The Structural Impossibility
A system cannot generate admissible evidence about its own compliance. This is not a legal presumption subject to rebuttal. It is a logical impossibility derivable from the structure of any audit system in which the auditor and the audited entity share the same write-path, the same administrative domain, or the same fault domain.
Formally: if audit substrate A and execution system E operate within fault domain D, then every record R produced by A is a function of E's control surface. A privileged adversary with access to D at firmware layer, hypervisor management plane, or through supply-chain insertion can deterministically shape R to reflect a desired state independent of the actual behavior of E. This condition is Probabilistic Circularity Bias (PCB).
PCB has a precise and irreversible forensic consequence: under adversarial cross-examination, any audit trail produced within a shared fault domain is challengeable as a product of the system it purports to document. The challenge does not require proof of actual manipulation. It requires only proof of the possibility of manipulation — which is structurally true for every endogenous system.
Courts of competent jurisdiction and regulatory supervisory authorities operating under NIS 2, DORA, and the AI Act are applying this challenge with increasing technical rigor in 2026. The question presented to organizations under examination is no longer whether an audit trail exists. The question is whether the audit trail could have been modified by the entity that produced it. For all endogenous systems, the answer is structurally yes.
1.2 The Hyperscale Primitive Inventory
Contemporary large-scale computing platforms have invested substantially and genuinely in security primitives that represent the state of the art in each of their respective domains:
Hardware-isolated secure processing environments operating within the main compute chassis, executing code and managing cryptographic material in a region physically protected from the host operating system and hypervisor layers.
Control-plane-isolated cryptographic sealing services that manage key hierarchies, perform signing operations, and enforce access policies through an administrative plane logically separated from the data plane.
High-granularity append-only event logging substrates that record API invocations, state transitions, and administrative actions at sub-millisecond resolution with cryptographic chaining between sequential entries.
Least-privilege identity and authorization systems enforcing fine-grained permission boundaries on every resource access, with immutable records of authorization decisions.
Non-intrusive observability pipelines collecting kernel-level, hypervisor-level, and hardware trace telemetry without requiring modification of observed workloads.
Long-term immutable archival storage with object-level cryptographic integrity verification and configurable retention lock policies.
This inventory is, in aggregate, the most sophisticated collection of security infrastructure ever made available at commercial scale. Each primitive is production-grade, operationally mature, and technically sound within its design scope.
Yet organizations operating on this infrastructure remain unable to produce court-admissible, regulatory-compliant evidentiary proof under active cross-examination. The reason is architectural, not technical. All primitives write their evidence into storage substrates that remain within the platform operator's administrative domain — the same domain as the system under audit. PCB is not eliminated by the sophistication of the individual primitives. It is reproduced at each tier of the stack.
The Core Diagnostic: The hyperscale platform has delivered all the parts. The parts have not been assembled into a forensically sovereign system. The missing element is not a component. It is the assembly doctrine that specifies the one external interface required to sever the endogenous write-path.
1.3 The Regulatory Demand in 2026
Five active EU regulatory frameworks collectively define a unified evidentiary standard that PCB prevents organizations from satisfying:
NIS 2 (Directive (EU) 2022/2555): supervisory authorities now require demonstration that security measures were operationally active at the moment of a significant incident, not reconstructed documentation produced after the fact. Post-hoc self-generated logs do not satisfy this requirement under examination.
DORA (Regulation (EU) 2022/2554): Article 17(3) requires financial entities to maintain contemporaneous ICT incident documentation sufficient for supervisory inspection. Contemporaneous is the operative constraint. An endogenous log cannot prove its own contemporaneity.
AI Act (Regulation (EU) 2024/1689): logging obligations for high-risk AI systems require records independent of the AI system's self-reporting. An AI system's own execution logs are the most acute form of PCB: the system is simultaneously the actor, the witness, and the record-keeper. Article 99 applies a three-tier penalty structure: 35M EUR or 7% of global annual turnover; 15M EUR or 3%; 7.5M EUR or 1%.
eIDAS 2 (Regulation (EU) 2024/1183): establishes the Qualified Trust Service Provider framework that defines legally recognized electronic timestamps and electronic seals applicable to the SOURCE 0 dual-timestamp mechanism.
EU Customs Trust and Check Reform: the Self-Release architecture requires authorized economic operators to prove, after automated goods release, that their declaration systems operated within approved parameters at the moment of release. This is a retroactive proof obligation that only a pre-execution evidence architecture can satisfy.
Across all five frameworks, the operative evidentiary standard is identical: proof must have been produced contemporaneously, by a system independent of the actor, anchored in a verifiable external trust chain. No endogenous audit system satisfies this standard.
2. Architectural Principles
The structural evolution and integrity constraints of the SOURCE 0 reference architecture are defined over a dynamic system state space tuple:
S_t = (G_t, CI_O(v)_t, HRD_t)
Where:
G_t = (V, E_t) represents the global system dependency graph evaluated over a discrete observation interval delta_t.
CI_O(v)_t represents the continuous Circularity Index evaluated relative to the fixed instrumentation oracle O_inst under active threat vector v.
HRD_t represents the immutable Historical Reality Dossier composite artifact committed to the external escrow endpoint.
To establish rigorous validation boundaries, the vertex set V (comprising virtual memory areas and kernel page table mappings) is partitioned into exactly three strictly disjoint, non-overlapping subsets:
V = P union E union A, where P intersect E = empty set, P intersect A = empty set, and E intersect A = empty set.
Subgraph P (Policy Domain): The system infrastructure containing human instructions, baseline configuration thresholds, and authorization parameters.
Subgraph E (Execution Domain - Tier 1): The runtime software environment where autonomous optimization loops, model inferences, and machine kinetic actions are performed.
Subgraph A (Evidence Domain - Tier 2): The hardware-isolated microkernel execution environment tasked with intercepting, sealing, and exporting compliance history.
The architecture enforces two core physical invariants across the system state space:
Invariant 1 — Spatial Decoupling: Subgraph A operates on a standalone terminal node sharing zero logical memory pages, CPU cache lines, or kernel page tables with Subgraph E. The separation is architectural and hardware-enforced by hardware resource partitioning, ensuring that any privileged adversarial mutation within domain E possesses zero write-path lineage over domain A, bounding the Measured Circularity Index below the threshold: CI_O(v)_t is less than or equal to epsilon, where epsilon < 0.05.
Invariant 2 — Temporal Sequence Gate: An ex-ante cryptographic execution blocking gate intercepts all communication lanes between Subgraph P and Subgraph E. The mechanism caches and blocks incoming parameter strings, withholding the execution release signal to the host processor until the deterministic state transition pipeline finishes and the resulting HRD_t file structure registration is confirmed by dual independent timestamp authorities.
2.2 The Dependency Entanglement Metric
The architecture computes a quantitative measure of fault domain contamination under active threat vector v, using an external hardware instrumentation substrate operating as the concrete implementation of the oracle profile:
CI_O(v)_t = ( |E_shared| - |E_noise| ) / ( |E_total| - |E_oracle_footprint| )
Where:
CI_O(v)_t: The continuous Circularity Index score at timestamp t under active threat vector v relative to the fixed instrumentation profile.
|E_total|: The absolute count of directed communication edges logged within the machine space during the discrete observation window delta_t.
|E_shared|: The explicit subset of directed edges representing shared virtual memory map addresses, page table entries, or kernel-level inter-process communication links crossing directly between Subgraph A and the state mutation channels controlled by P or E.
|E_oracle_footprint|: The known memory execution footprint generated natively by the telemetry capture framework runtime, mathematically subtracted to prevent self-referential trace amplification.
|E_noise|: The statistical variance edge factor, hypervisor scheduling jitter, and structural propagation delay introduced by the physical data relay hardware (the relay latency correction component).
2.3 The System Acceptance Theorem
The architectural integrity and functional validity of the reference architecture within a live operational system are governed by a strict logical conjunction. The system state is defined as structurally verified if and only if all of the following tracking conditions are simultaneously satisfied:
SOURCE 0 = VALID if and only if ( for all v in V_attacks, CI_O(v)_t is less than or equal to epsilon AND Evaluation(HRD) is less than Strata 3 )
Where the operational verification boundaries require compliance across exactly five structural parameters:
Parameter 1 (Circularity Boundary): For all active threat vectors v within the adversarial set V_attacks, the dynamic Circularity Index score must satisfy: CI_O(v)_t is less than or equal to epsilon, where the empirical leakage ceiling is fixed as a constant equal to 0.05.
Parameter 2 (Artifact Integrity): The separate Evaluation Function parsing the Historical Reality Dossier must return an un-broken confidence score matching Strata 1 or Strata 2, mathematically ensuring that Evaluation(HRD) is less than Strata 3.
Parameter 3 (Temporal Alignment): External escrow vault deposit acknowledgment is confirmed by two independent, synchronized clock sources within the calibrated pre-execution window: Effective_delta_t.
Parameter 4 (Trust Root Exogeneity): The cryptographic public key infrastructure and master signing root used to verify hardware enclave attestation are managed entirely outside the platform operator's administrative domain.
Parameter 5 (Sequence Monotonicity): The hardware-enforced monotonic event counter embedded inside the physical unidirectional data relay is verified as strictly increasing across all sequential observation intervals, proving the absence of frame-suppression or trace-omission actions.
If the architecture violates any single element of this conjunction under runtime testing or real-time metric collection, the system state transitions instantly to a terminal REJECTED failure domain.
2.4 The Epistemic Boundary
The epistemic boundary of the SOURCE 0 reference architecture is precise, non-negotiable, and structurally bounded. The execution of the deterministic sealing pipeline establishes a clear operational demarcation line between internal system authorization and external reality:
Fact 1 (The Domain of Proof): The Historical Reality Dossier (HRD) seals the exact state of human managerial diligence at instant T-0. It constitutes cryptographically anchored, unalterable proof of what the automation infrastructure was explicitly instructed to execute, at what verified atomic timestamp, and under what attested microarchitectural configuration states, strictly prior to runtime machine execution.
Fact 2 (The Domain of Exclusion): The architecture does not evaluate, log, or certify the factual accuracy or objective truth of the external world-states that preceded T-0. It documents the mandate issued to the machine; it does not validate the external inputs provided by the operator.
This structural boundary condition does not constitute an operational limitation or an engineering vulnerability. It represents the framework's primary legal protective property, establishing the precise parameters for non-binding cross-border mapping templates.
An organization that correctly limits its forensic defense to this boundary occupies an analytically defensible and verifiable evidentiary position under active regulatory cross-examination. Conversely, any deployment model that overstates this property by claiming to validate external real-world correctness exposes its entire compliance stack to a structural refutation condition under hostile evaluation.
3.1 Vulnerability I: Observability Substrate Co-Residency
Description: The continuous circularity metric tracking domain collapse is computed by a local hardware instrumentation substrate (O_hw) using kernel-level event tracing and hardware processor trace loops. In its baseline configuration, O_hw remains physically co-resident within the virtual memory address spaces and hypervisor scheduling frames of Subgraph E—the active execution plane it is tasked to monitor.
Attack Model: A privileged adversary with root-level escalation over the host kernel or hypervisor management plane can suppress or selectively filter telemetry events before they reach the trace buffers; modify processor execution trace logs directly inside volatile memory maps prior to export; or artificially manipulate the metric computation to force the observed circularity score below the epsilon ceiling under active system compromise. The System Acceptance Theorem is thereby satisfied by a compromised environment that has co-opted its own instrumentation layers. This constitutes second-order Probabilistic Circularity Bias (PCB) operating on the metric designed to detect first-order PCB.
Correction: The local instrumentation substrate (O_hw) is uncoupled from metric analysis and re-routed to an External Measuring Unit (O_ext). Telemetry from Subgraph E is pushed to O_ext exclusively through a hardware-enforced unidirectional data relay (diode network topology) sharing zero feedback logic or volatile registers with the host compute chassis. The Circularity Index equation is updated to integrate the structural propagation delay and hypervisor scheduling jitter into a single calibrated noise variable: |E_noise|. An adversary can suppress events but cannot inject false execution edges into O_ext; any unauthorized trace suppression is instantly detected via a hardware-locked monotonic sequence counter embedded inside the unidirectional relay lane.
Platform Primitive Mapping: High-granularity, non-intrusive processor execution trace substrates and asynchronous kernel-level telemetry streaming loops provide the collection fabric. Routing their raw data blocks through a hardware unidirectional data relay before ingestion by the external analysis node requires zero modifications to the core workload execution pipelines.
3.2 Vulnerability II: The delta_t Temporal Leak
Description: The pre-execution temporal gate delta_t is operationalized as a strict blocking sequence: no instruction variable is cleared for execution inside Subgraph E before a verified Historical Reality Dossier (HRD) deposit is committed to the external escrow endpoint. Under high-throughput enterprise workloads, three distinct forces create probabilistic violations of this constraint: network packet delivery jitter on the escrow export interface, hypervisor scheduling backpressure that preempts the telemetry-sealing node execution mid-transit, and clock drift between the host compute chassis and the isolated verification terminal.
Attack Model: The synchronization variance between the execution plane and the sealing gate creates a temporal leak window where the execution release signal is issued to the host processor before external vault acknowledgment is completed. This degradation transforms the ex-ante timeline gate from a structural invariant into a probabilistic property. A single instance of machine execution preceding external vault confirmation invalidates the forensic integrity of the HRD artifact under cross-examination.
Correction: The pre-execution blocking sequence gate is hardcoded onto a dedicated hardware interrupt controller running a physical clock reference completely isolated from hypervisor scheduling preemption. A temporal calibration calculation is introduced to bound the processing window: Effective_delta_t = delta_t - tau_drift - tau_network_jitter_ceiling. Effective_delta_t must satisfy the boundary condition: Effective_delta_t > 0. The variable tau_drift is monitored continuously against an exogenous stratum-1 time reference independent of any host-managed network time protocols (NTP). The variable tau_network_jitter_ceiling is locked at the 99.9th percentile of observed round-trip latencies over a rolling 30-day profiling baseline. The escrow vault must return a signed acknowledgment token bound to its own separate clock infrastructure. Final execution release requires both the isolated gate clock and the external vault clock to confirm verification within tau_drift tolerance. A failure to clear this dual-clock handshake within Effective_delta_t triggers immediate hardened suspension: execution lanes are blocked, a high-priority alert is generated, and the incomplete HRD fragment is locked in hardware-secured read-only media for forensic recovery.
Platform Primitive Mapping: High-granularity append-only event logging substrates with cryptographic ledger chaining provide the transaction serialization. The required structural addition is the integration of an exogenous stratum-1 clock reference and a dedicated hardware interrupt controller on the isolated terminal node.
3.3 Vulnerability III: Attestation Chain Endogeneity
Description: The hardware-isolated secure enclave primitive within the Evidence domain (Subgraph A) generates an asymmetric remote attestation quote confirming the cryptographic identity and unmodified microcode state of the sealing engine. Verification of this attestation report requires navigating a chain of trust terminating at a root public key certificate authority. If that master root certificate is hosted or managed inside the platform operator's administrative domain, the attestation infrastructure is endogenous.
Attack Model: A platform operator subject to unauthorized administrative coercion, or an adversary who has achieved privilege escalation over the platform operator's internal certificate management infrastructure, can forge valid hardware attestation reports for a modified or compromised enclave configuration. The cryptographic verification step—designed to decouple the HRD from the host platform's fault domain—remains dependent on a trust anchor residing inside the adversary's compromise perimeter. Endogeneity is reinstated through the trust hierarchy rather than the data write-path.
Correction: The attestation verification chain for the secure enclave must terminate at a root public key managed by an entity possessing zero contractual, operational, or logical relationship to the host platform operator. Admissible roots are strictly restricted to eIDAS 2 Qualified Trust Service Providers, sovereign national public key infrastructures, or statutory judicial officers. The dual-timestamp architecture is extended to sign and freeze the remote attestation quote itself at instant T-0, rendering the enclave state verifiable by any external third-party auditor using public trust lists, completely bypassing the platform operator's identity domain.
Platform Primitive Mapping: Hardware-rooted isolated secure enclaves executing memory page encryption provide the isolated signing environment. The correction requires a configuration change to route the attestation validation pathway through an exogenous qualified root certificate authority, avoiding architectural redesign of the native CPU silicon.
4. Implementation Blueprint
4.1 Architecture Overview
The corrected SOURCE 0 architecture instantiates three distinct tiers characterized by non-overlapping security properties, interface boundaries, and explicit evidentiary roles. Implementation on standard hyperscale infrastructure requires zero infrastructural redesign, zero novel cryptographic mechanisms, and zero modification to existing compliance software tracking. The deployment is bounded exclusively by one public legal partnership and two external web API integrations.
4.2 Tier 1: Execution Plane
Role: The operational compute layer where automated, distributed workflows are processed, including agentic AI instruction pipelines, autonomous financial transactions, hazardous logistics corridor asset allocation, and customs declarations generated under automated self-release frameworks.
Evidentiary Status: No forensic weight or probatory value is claimed for transactional logs or telemetry artifacts produced natively within Tier 1. Its outputs are treated strictly as operational mutations. The independent, unalterable proof of its authorized input parameters is stored exclusively inside Tier 3.
Constraint: Every operational instruction routed to Tier 1 must be systematically intercepted and released by the Tier 2 pre-execution sequence gate. This constitutes the sole boundary constraint imposed on the live execution engine. Existing services, container clusters, and host architectures operate without internal modification.
Platform Primitive: Any general-purpose compute infrastructure available on standard cloud platforms.
4.3 Tier 2: Pre-Execution Evidentiary Gate
Role: An isolated microkernel gate intercepts every parameter string targeting Tier 1 within a pre-execution window. The gate executes on a standalone processing terminal running a minimal trusted computing base, maintaining absolute spatial decoupling from the host operating system.
Processing Sequence:
Instruction parameter ingestion on the isolated Evidence terminal via an authenticated, inbound-only hardware transmission channel.
Structural canonical serialization using the deterministic JSON Canonicalization Scheme (IETF RFC 8785), stripping whitespace and sorting dictionary keys alphabetically to ensure absolute byte-for-byte consistency across separate verification nodes.
Computation of a fixed-length cryptographic signature via an unsalted secure hashing algorithm conforming to FIPS 180-4 (SHA-256). The total omission of cryptographic salt is required to support offline reproducibility, enabling any third-party auditor in possession of the raw input text to replicate the hash exactly without accessing protected keysspace variables.
Transmission of the hash to a local hardware-rooted secure enclave primitive utilizing memory encryption keys, generating an asymmetric remote attestation report signed directly by CPU microcode, terminating at an exogenous qualified root.
Parallel execution of dual cryptographic timestamp requests targeting two separate, independent Qualified Trust Service Providers conforming to the IETF RFC 3161 protocol, binding both the payload hash and the enclave microcode attestation report.
Monotonic grouping of the canonical instruction, the enclave attestation quote, and the dual qualified timestamp tokens into a single, structured Historical Reality Dossier (HRD) file object.
Encrypted transmission of the HRD artifact to the Tier 3 External Judicial Vault via a dedicated network interface lane configured with a physical unidirectional diode topology to block hypervisor backpressure loops.
Clearance and release of the execution signal to Tier 1, triggered exclusively when an authenticated vault deposit acknowledgment token is validated by both the Evidence terminal's isolated internal clock and the vault's independent clock infrastructure within the window: Effective_delta_t.
Failure Behavior: If vault acknowledgment is not parsed or fails confirmation within Effective_delta_t, the Evidence terminal executes a hardened suspension routine: the execution release signal is permanently withheld from Tier 1, a high-priority system exception alert is pushed to risk operators, and the incomplete HRD fragment is instantly written to local hardware-locked read-only storage media attached to the terminal physical bus for forensic recovery. This suspension state cannot be overridden, bypassed, or cleared by any compute process running inside Tier 1 or by any administrative role within the platform operator's identity namespace.
Platform Primitives: Hardware-rooted isolated secure enclaves with active cryptographic memory encryption; control-plane-isolated cryptographic sealing services; high-granularity append-only event logging substrates with ledger chaining; and fine-grained least-privilege access management controls configured such that no identity governing Tier 1 holds read or write entitlement inside Tier 2.
4.4 Tier 3: External Judicial Vault
Role and Custodian: An append-only, structurally immutable archive of HRD artifacts, hosted entirely outside the administrative and network domain of both the platform operator and the organization under audit. Custody, verification, and key management are governed by a Commissaire de Justice—a statutory judicial officer appointed by public legal authority and recognized across all EU member states under the Brussels I Recast Regulation (Regulation No 1215/2012). The custodian does not operate as a third-party commercial software vendor, but functions as an officer of the public judicial system empowered to issue legally sovereign custody records admissible as full right evidence (preuve de plein droit) before any competent regulatory authority or court of law.
Structural Properties:
Complete network, logical, and administrative isolation from the cloud platform account infrastructure.
Chain of custody traceable directly to a named, legally accountable public officer.
Vault access logs and entry signatures subject to automated verification checks by an independent verifier consensus framework.
Serialized HRD entries bound sequentially using the same hardware-enforced monotonic counter implemented at the Tier 2 unidirectional relay.
Retention lock and cryptographic deletion policies controlled exclusively by the statutory custodian under public legal mandating rules, preventing premature deletion by the organization or the platform operator.
Platform Primitive: Long-term immutable archival storage featuring object-level cryptographic integrity validation and immutable write-once retention locks, where configuration access keys are held under the sole custody of the external judicial authority.
Judicial Opposability: Any supervisory body or court of competent jurisdiction can verify the complete chain of custody by requesting direct production of the HRD artifacts from the Commissaire de Justice. The resulting documentation satisfies the explicit chain-of-custody standards required by international digital forensic and procedural evidence frameworks, eliminating the requirement to provide external investigators with raw administrative access to the organization's internal computing networks.
4.5 Integration Architecture: The Two-Call Interface
The execution of the SOURCE 0 reference architecture requires no internal modification to existing production databases or application wrappers, bounding the integration footprint to exactly two HTTPS API call invocations per sealed instruction inside the pre-execution time window:
Call 1 — Dual-Timestamp Issuance:
Endpoints: Two separate, geographically isolated IETF RFC 3161 qualified timestamp authorities operated by independent Qualified Trust Service Providers.
Payload: The SHA-256 hash of the canonicalized parameter string concatenated with the active secure enclave remote attestation report.
Response: Two distinct cryptographically signed timestamp tokens.
Latency: Less than 200 milliseconds processed in parallel under standard network conditions.
Call 2 — Vault Export and Acknowledgment:
Endpoint: The secure API endpoint of the Commissaire de Justice external judicial vault.
Payload: The consolidated HRD file structure object containing the canonical parameters, the hash digest, the enclave attestation report, and the dual qualified timestamp tokens.
Response: An asymmetric, legally sovereign vault acknowledgment token bound to the vault's independent clock infrastructure.
Latency: Less than 500 milliseconds under standard operating conditions.
The cumulative processing latency floor is bounded below 700 milliseconds per invocation under standard parameters. For high-frequency, millisecond-scale transactional environments where individual instruction latency constraints prevent per-instruction network calls, the gate executes in Batch Sealing Mode. A batch of up to 10,000 instructions is canonicalized and compiled into a cryptographic Merkle tree structure inside the secure enclave; the single resulting Merkle root hash is signed, dual-timestamped, and exported to the judicial vault in a single processing cycle. The chronological provenance and integrity of any individual instruction within the batch remain completely verifiable by tracing its matching Merkle path proof, scaling effective throughput up to 10,000 instructions per 900 milliseconds per gate instance without expanding the core execution attack surface.
Integration Cost Summary: One public legal partnership agreement with a Commissaire de Justice; two qualified timestamp authority registrations; two external HTTPS API endpoints; and one configuration mapping to route enclave attestation signatures through an exogenous qualified certificate root. Zero platform structural redesign, zero novel hardware, and zero new compliance tracking software. Time-to-production: weeks.
5. Regulatory Alignment
5.1 NIS 2 (Directive (EU) 2022/2555)
Article 21(2) requires operators of essential and important entities to implement measures for incident handling, business continuity, and supply chain security. NIS 2 supervisory authorities increasingly require demonstration that measures were operationally active at the moment of a significant incident, not reconstructed in its aftermath.
The HRD provides the pre-execution instruction record that proves policy enforcement preceded the incident. The sealed Tier 3 vault entry constitutes contemporaneous proof of the system's authorized instruction state at T-0, produced and externally anchored before the event, independent of the system's post-incident self-reporting.
5.2 DORA (Regulation (EU) 2022/2554)
Article 17(3) requires financial entities to maintain contemporaneous ICT incident documentation sufficient for regulatory reporting and supervisory inspection. The operative constraint is contemporaneous: documentation produced after the fact does not satisfy this requirement under adversarial examination.
The Tier 3 vault entry is contemporaneous by construction: produced, sealed, and dual-clock-confirmed before the execution it documents. The Commissaire de Justice provides the chain of custody documentation required for regulatory production to competent authorities.
Note: DORA's scope is defined in Article 2 and applies to financial entities. Organizations outside this scope operate under NIS 2 or sector-specific frameworks.
5.3 AI Act (Regulation (EU) 2024/1689)
The AI Act imposes logging obligations on providers and deployers of high-risk AI systems requiring records independent of the AI system's self-reporting. Agentic AI execution — autonomous instruction chaining without per-instruction human approval — represents the most acute instance of PCB in current deployments: the AI system is simultaneously the actor, the instruction generator, and the sole record-keeper of its own behavior.
The SOURCE 0 pre-execution gate applied to agentic AI instruction chains produces a log that is structurally external to the AI system. Instructions are sealed before they are executed, by a system with no dependency on the AI system's output. The HRD is independent of the AI system's self-reporting in the sense required by Article 99 conformity assessment obligations. The three-tier penalty structure (35M EUR / 7%; 15M EUR / 3%; 7.5M EUR / 1%) applies to high-risk system compliance failures; a pre-execution evidentiary architecture is the most direct mitigation of the highest penalty tier.
5.4 eIDAS 2 (Regulation (EU) 2024/1183)
eIDAS 2 establishes the legal framework within which the SOURCE 0 trust anchoring operates. The dual-timestamp mechanism (RFC 3161, two independent QTSPs on national eIDAS trust lists) provides qualified electronic timestamps with the legal effects defined under eIDAS 2 for electronic evidence purposes. The exogenous attestation root anchored in a QTSP-managed certificate hierarchy establishes enclave integrity state under a trust framework legally recognized in all EU member states and in third countries with EU digital trust equivalence arrangements.
5.5 EU Customs Trust and Check Reform
The Trust and Check / Self-Release architecture introduces a regime in which authorized economic operators release goods before formal customs clearance, based on pre-authorized declaration frameworks. The post-release regulatory obligation — to prove that the automated declaration system operated within approved parameters at the moment of release — is structurally unsatisfiable by post-hoc self-generated logs.
SOURCE 0 converts this post-hoc proof obligation into a pre-execution architectural guarantee. Every declaration instruction is sealed before transmission. The HRD entry proves the content and timing of the authorized instruction independently of the declaration system's own records, satisfying the Trust and Check evidential standard without requiring any modification to the declaration system's operational architecture.
6. Strategic Impact
6.1 The Competitive Architecture of Evidentiary Sovereignty
The EU regulatory environment of 2026 has created a new competitive dimension in cloud computing: evidentiary sovereignty. Organizations operating under the three-tier architecture occupy a structurally different legal position than those operating under endogenous audit systems. They can produce, for any automated decision event, a sealed instruction record that predates the event, was produced by a system external to the event system, is anchored in a trust chain independent of any administrative domain, and is held in statutory judicial custody.
This position is not replicable by endogenous means. It is replicable only through the SOURCE 0 assembly doctrine. A hyperscale platform that integrates this doctrine as a native service offering acquires a capability that no competing platform currently provides: the ability to deliver EU-compliant evidentiary sovereignty to its customers as a managed architecture, rather than requiring customers to construct it independently.
The platform that first operationalizes this assembly — integrating its existing primitive stack with the two external API interfaces required — establishes a structural lead in the EU regulated market that compounds with each additional regulatory enforcement action taken against organizations operating under endogenous systems.
6.2 The Liability Profile Transformation
An organization operating without SOURCE 0-class evidentiary governance must prove its systems behaved correctly using records produced by those same systems. Under PCB, this proof is challengeable as circular and can be neutralized by any competent adversarial examiner. The liability exposure is structurally unbounded in any proceeding where the opposing party is technically sophisticated.
An organization operating under the three-tier architecture can produce, on demand, a sealed HRD entry for any automated decision event that: predates the event by a measurable and verifiable interval confirmed by two independent clocks; was produced by a system external to the event system with zero shared resources; is anchored in an attestation chain terminating at a QTSP root; and is held in statutory custody by a judicial officer empowered to produce it as evidence. The burden of proof shifts from assertion to demonstration.
6.3 Underwriting and Insurability
The insurability of autonomous AI execution risk is currently constrained by the absence of a pre-execution evidentiary anchor. A technical underwriter covering an organization's agentic AI deployment cannot determine, from the organization's own records, the boundary between authorized and unauthorized machine behavior at the moment a loss event occurred. The endogenous audit trail is the only available reference, and it is structurally challengeable.
SOURCE 0 provides the technical underwriting anchor. For any claimed event, the carrier can request production of the HRD entry from the Commissaire de Justice. The entry specifies the precise instruction state at T-0, the attested environmental conditions, and the externally anchored timestamp, all produced before the execution that generated the loss. The underwriting question shifts from policy language interpretation to architectural verification: does the insured operate under a documented three-tier evidentiary architecture with an active judicial vault? This is a binary verifiable question. Binary verifiable questions are insurable.
6.4 Re-Federation Audit Criteria
Organizations and platforms evaluating SOURCE 0 compliance should apply the following six criteria:
Criterion 1 — Fault Domain Separation: The audit substrate cannot be written to, modified, or deleted by any process, account, or administrative role with write access to the operational system under audit.
Criterion 2 — Pre-Execution Sequencing: For any automated decision, the evidence artifact must demonstrably predate the execution it documents, verified by externally anchored timestamps from independent clock sources.
Criterion 3 — Attestation Chain Exogeneity: The trust chain for cryptographic signing of evidence artifacts must terminate at a root certificate managed outside the platform operator's administrative domain.
Criterion 4 — Hash Reproducibility: Any party in possession of the original instruction must independently reproduce the hash value recorded in the HRD without access to any secret material.
Criterion 5 — Vault Independence: The External Judicial Vault must be subject to independent audit, with its own monotonic event log verified by a party external to both vault operator and organization.
Criterion 6 — Judicial Opposability: The HRD custody chain must be managed by a statutory officer with documented legal standing to produce the record in proceedings before courts and regulatory authorities in all relevant EU jurisdictions.
7. Conclusion
Evidentiary circularity is not a security flaw or a configuration gap within any isolated platform primitive. It is an unavoidable structural consequence of assembling those primitives without the overarching engineering doctrine required to dictate their forensic and legal validity. Every hyperscale cloud platform that has heavily invested in the deployment of hardware-isolated secure enclaves, control-plane-isolated cryptographic vaults, high-granularity append-only transaction logs, non-intrusive hypervisor-level observability pipelines, and object-level immutable archival storage has already successfully compiled approximately 95% of the physical components necessary to build a legally defensible forensic tracking infrastructure. However, none has assembled them into a unified, non-hybrid system.
The missing 5% does not consist of novel computing infrastructure, untested hardware, or un-reconciled encryption math. It is bounded strictly by two external, out-of-band HTTPS API interface connections: one routed to a pair of separate eIDAS 2 Qualified Trust Service Providers for continuous dual-timestamp token issuance, and one routed directly to a Commissaire de Justice for independent external judicial vault custody. The Commissaire de Justice is not a speculative legal concept or a newly introduced regulatory framework. This statutory public officer is operational, acts EU-wide under the cross-border mandates of the Brussels I Recast Regulation (Regulation No 1215/2012), and holds the unique public authority required to issue legally sovereign custody records that are incontestable and admissible before any EU member state court or regulatory supervisor.
The SOURCE 0 assembly doctrine specifies exactly how the existing primitive inventory and the two external interfaces must be ordered, physically separated, and temporally sequenced to eliminate the Probabilistic Circularity Bias (PCB) and block the occurrence of observability collapse. By enforcing strict spatial and temporal boundaries, the architecture satisfies the rigorous contemporaneous evidence requirements shared across NIS 2, DORA, the AI Act, eIDAS 2, and the EU Customs Trust and Check Reform. The system model is closed. The hardware primitives are active on production clouds. The external judicial anchor is accessible. The regulations mandating this standard are fully enforceable.
The sole remaining deployment variable is the organizational decision to integrate. For any enterprise or cloud platform possessing the core primitive inventory, the operational path to full deployment is constrained exclusively by a strategic partnership agreement, two external API registrations, and a configuration update to the remote attestation root routing keyspace. The timeline to live production is measured in weeks.
The European regulated infrastructure market of 2026 is executing a violent selection process in favor of complete evidentiary sovereignty. The organizations and computing platforms that choose to operationalize this selection criterion first will define the compliance baseline for the regulatory epoch that follows. The SOURCE 0 reference architecture constitutes the plan de montage manquant—the missing assembly blueprint. Every component required to execute it is already built and resting inside the silicon.
8. Partnership Architecture and Performance Baseline
8.1 Throughput Baseline: Tier 2 Gate Performance
The following performance parameters are derived from reference implementation modeling on isolated microkernel hardware with standard EU network connectivity to eIDAS 2 Qualified Trust Service Provider (QTSP) endpoints. They are provided to enable internal business case construction by technical and commercial evaluators.
Sequential gate processing metrics per isolated instruction:
Parameter 1 (Canonicalization Sequence): Algorithmic canonical formatting under IETF RFC 8785 executes in less than 1 millisecond on standard compute hardware for instruction payloads under 64KB.
Parameter 2 (Cryptographic Hashing): Unsalted cryptographic hash computation under FIPS 180-4 (SHA-256) executes in less than 1 millisecond.
Parameter 3 (Hardware Sealing): Asymmetric hardware secure enclave signing operations require between 5 and 15 milliseconds, scaling deterministically with enclave generation and private keyspace allocation parameters.
Parameter 4 (Distributed Synchronization): Dual QTSP network timestamp tokens issuance via parallel IETF RFC 3161 invocations requires between 100 and 200 milliseconds per provider under standard EU network paths, executed in parallel and bounded exclusively by the latency of the slower responding root.
Parameter 5 (External Escrow Entry): Vault data replication and acknowledgment receipt from the Commissaire de Justice api endpoint requires between 300 and 500 milliseconds under standard operating parameters.
The cumulative processing latency floor is bounded between 400 and 700 milliseconds per discrete instruction under normal operating conditions. The effective system throughput in strict sequential mode maps between 1.4 and 2.5 instructions per second per gate instance. In a parallel gate deployment involving N concurrent processing gates, the system throughput scales linearly to N x 1.4 to N x 2.5 instructions per second, featuring zero architectural ceiling on the expansion value of N.
For high-frequency, automated enterprise workflows where individual instruction latency constraints prevent per-instruction network loops, the gate executes in Batch Sealing Mode. A transaction batch of up to 10,000 instructions is canonicalized, compiled into a cryptographic Merkle tree structure inside the secure enclave, signed, dual-timestamped, and exported to the external judicial vault in a single processing cycle. The historical tracking integrity of any individual instruction within the batch remains completely verifiable via offline Merkle path proof extraction, bypassing the requirement for isolated gate cycles. The cumulative batch gate processing latency is bounded between 600 and 900 milliseconds regardless of batch volume density up to 10,000 instructions, scaling effective throughput up to 10,000 instructions per 900 milliseconds per gate instance.
The local infrastructure footprint per active gate instance is restricted to: one isolated microkernel terminal node, one hardware interrupt controller, one exogenous stratum-1 clock reference, and one physical unidirectional hardware data relay. The stack is deployable as a dedicated physical hardware appliance or as a hardware-isolated partition on existing bare-metal infrastructure, subject to the strict spatial separation invariants defined in Section 2.1.
These performance vectors establish that the SOURCE 0 architecture is operationally viable for high-throughput automated environments, including agentic AI instruction pipelines, automated customs classification networks, and financial transaction authorization systems. The gate mechanism does not introduce an operational throughput ceiling; it introduces an isolated latency floor that is architecturally bounded and operationally predictable. Performance parameters represent reference implementation modeling. Calibration against specific hardware and network configurations is performed during the private briefing phase.
8.2 Partnership Structure: Three Available Models
The commercial and technical integration of the SOURCE 0 doctrine is organized across three distinct partnership models, designed to align with the specific infrastructure topology of the platform operator:
Model A — Doctrine Licensing: The platform operator licenses the complete SOURCE 0 assembly doctrine, the closed-form system specification (the system state space equation, the continuous Circularity Index metric, the System Acceptance Theorem, and the three-tier topology), and the six Re-Federation Audit Criteria to serve as the baseline evidentiary governance standard for a managed cloud offering. The platform operator engineers and instantiates the Tier 2 gate and Tier 3 vault interface blocks using internal development resources, subject to explicit conformance verification audits. Cabinet Jean-François ELSEN / SRL AXIOTRANS provides independent conformance review, continuous doctrinal updates as EU regulatory frameworks evolve, and expert testimony support during judicial or regulatory cross-examination. The platform operator establishes the Commissaire de Justice judicial vault partnership independently, with initial operational facilitation provided by Cabinet Jean-François ELSEN / SRL AXIOTRANS.
Model B — OEM Integration: The platform operator integrates the SOURCE 0® Tier 2 gate as a pre-validated reference implementation blueprint, deployed as a co-branded or white-labeled managed service within the platform's existing security and compliance catalog. Cabinet Jean-François ELSEN / SRL AXIOTRANS delivers the functional reference implementation specifications, the Commissaire de Justice external judicial vault partnership as a pre-bundled structural component, and continuous doctrinal system governance. The platform operator manages deployment logistics, enterprise customer onboarding, and commercial distribution layers within its established enterprise agreement frameworks. This model is optimized for cloud infrastructure providers seeking immediate time-to-market deployment without internal engineering overhead, mapping directly onto the six production primitives identified in Section 1.2 via the two-call external interface API configuration described in Section 4.5.
Model C — Strategic Co-Commercialization: The platform operator and Cabinet Jean-François ELSEN / SRL AXIOTRANS establish a joint go-to-market commercial structure targeting highly regulated enterprise organizations operating within the EU financial services, critical logistics infrastructure, maritime transport corridors, and public sector verticals. SOURCE 0 is deployed as the foundational evidentiary governance and sovereignty layer of the platform's EU regulatory offering, featuring the Commissaire de Justice external judicial vault as a jointly marketed infrastructure service component. Commercial revenue allocation, mutual exclusivity terms, and territorial operational boundaries are governed by private bilateral negotiation under non-disclosure agreement (NDA) constraints prior to public release. This model is optimized for hyperscale platform operators seeking a structural, non-replicable competitive differentiator within the EU regulated space for the complete duration of the active regulatory epoch.
8.3 Engagement Protocol: From Public Document to Private Conversation
This whitepaper serves as an authoritative public release specification designed to establish the logical model, mathematical primitives, and hardware constraints of the SOURCE 0 architecture for qualified systems engineers and risk evaluators.
For cloud platform operators, tier-1 systems integrators, technical risk underwriters, and regulatory compliance counsel who have verified a specific integration use case or deployment tract, an independent private briefing package is accessible under non-disclosure agreement (NDA) constraints. The private briefing includes:
A reference implementation specification for Tier 2 gate deployment calibrated to the specific microarchitectural and primitive stack of the requesting platform.
A draft commercial term sheet mapping the selected partnership model parameters.
A Commissaire de Justice operational brief detailing vault storage terms, cross-border EU jurisdictional mechanics, and chain of custody documentation standards.
A customized regulatory impact scenario mapping the active EU frameworks onto the organization's explicit vertical compliance risks.
A continuous performance modeling matrix calibrated to the organization's targeted instruction volume, parallel gate clustering requirements, and latency thresholds.
The private technical briefing sequence is initiated via a single contact endpoint through jfelsen.com. The onboarding track bypasses public RFP loops, multi-vendor evaluation processes, and open-tender cycles, remaining confidential until both parties explicitly elect otherwise.
The first-mover market entry window for sovereign evidentiary architecture is constrained by the enforcement calendars of the active EU regulatory frameworks. NIS 2 supervisory interventions, DORA competent authority inspections, and AI Act market surveillance audits are actively generating formal evidence production demands. The organizations that are forced to defend a circular, self-generated software log before they have deployed an ex-ante decoupling architecture will define, by negative example, the operational urgency that their direct competitors will execute upon.
First contact: jfelsen.com.
Contact and Engagement
Cabinet Jean-François ELSENSenior Forensic Auditor | Judicial Specialist in Digital Evidence | ADR Safety Advisor (CSTMD, Explosives Class 1)Brussels / Charleroi, Belgium |jfelsen.com
For architectural assessment under the SOURCE 0 diagnostic protocol, partnership structuring with a Commissaire de Justice, or technical underwriting integration: contact throughjfelsen.com.
Regulatory and Technical References
NIS 2: Directive (EU) 2022/2555 of the European Parliament and of the Council
DORA: Regulation (EU) 2022/2554 of the European Parliament and of the Council
AI Act: Regulation (EU) 2024/1689 of the European Parliament and of the Council
eIDAS 2: Regulation (EU) 2024/1183 of the European Parliament and of the Council
CER Directive: Directive (EU) 2022/2557 of the European Parliament and of the Council
Brussels I Recast: Regulation (EU) No 1215/2012 of the European Parliament and of the Council
RFC 3161: Internet X.509 Public Key Infrastructure Time-Stamp Protocol (TSP)
RFC 8785: JSON Canonicalization Scheme (JCS)
FIPS 180-4: Secure Hash Standard (SHS) — SHA-256
MITRE ATT&CK: Techniques T1056 (Input Capture) and T1059 (Command and Scripting Interpreter) — referenced for threat modeling of in-machine observability substrate vulnerabilities