EVIDE Visual Doctrine

AI Act Requires Logs. Logs Are Not Evidence.

The EU AI Act requires logging and traceability under Articles 12, 14, 16, 17, and 26. But logs remain internally controlled records generated and retained by the same organization that made the decision. This article introduces the concept of same-party evidence and explains why execution logs alone cannot independently reconstruct responsibility conditions during disputes, audits, or litigation.

Same-Party Evidence

The organization investigating itself through its own logs. When evidence is produced, retained, and interpreted by the same party whose decisions it is meant to document, independence collapses before the dispute even begins. This article establishes same-party evidence as a pillar concept of evidentiary AI governance - and introduces the four objections organizations raise, and why none of them resolves the structural independence problem.

Reconstruction Independence

When evidence depends on the same system that generated the dispute, reconstruction is not independent. This article maps the architectural requirement for externally reconstructable records - evidence that can be verified without access to the originating system, its operators, or its runtime context. It introduces the Reconstruction Test and the four dissolution conditions that progressively erode internal evidence.

Why Execution Evidence Is Not Enough

"What happened" is not the same as "who stands behind the outcome." This article maps the structural difference between execution, accountability, and reconstructability - three different evidentiary questions that require three different answers. It introduces the asymmetry: governance and reconstructability do not fail together.

The Missing Evidentiary Property

Four articles. Four different contexts. Four different vocabularies. This article shows they were never describing four different problems - they were describing the same missing property, observed from four different angles. It names that property for the first time: Evidentiary Independence, the root condition beneath same-party evidence, reconstruction independence, execution evidence, and governance asymmetry.

EVIDE Is Born

Article 05 ended with a question. Can evidentiary independence be designed? This article begins with the answer. Not a product description - the architecture that becomes inevitable once a property is named. External anchoring, decision-time, evidentiary neutrality, independent persistence: the principles a system must satisfy before it can preserve evidentiary independence at all.

What EVIDE Actually Preserves

EVIDE does not certify decisions. It preserves the conditions under which decisions were made and responsibility was exercised. Not four independent categories, but one accountability relationship examined from four angles: governance state, authority conditions, oversight state, admissibility markers. This article maps what is preserved - and, with equal precision, what is deliberately left outside that boundary.

Crossing-Time

Evidentiary independence cannot emerge after a dispute. It must exist before the dispute becomes possible. This article defines crossing-time - the architectural moment a decision's accountability conditions must already be fixed - and the three states that describe how much of that relationship survived it: candidate, verified, and unverified.

The Evidentiary Deposit

From governance state to evidentiary object.

Once a governance state is externally anchored, it becomes a self-contained object. This article opens that object: what an evidentiary deposit contains, why each element it contains was made necessary by principles already established in the series - and, just as importantly, what it deliberately does not contain.

Accountability Closure

Custody without control - not who holds the deposit, but what holding it does not permit. Verification fixes the deposit; closure fixes the relationship between that deposit and everyone who may later need to examine it. This article closes Phase II by defining the architectural event at which an organization stops determining whether, and how, its own evidence can be examined.

FCC: Forensic Cross-Check

A deposit can satisfy every condition this series has established and still contain a relationship that does not hold. FCC cross-checks the declared classification against the declared runtime visibility that was meant to support it - producing four states, none of them a verdict: stable, degraded, unknown, broken. The first mechanism of Phase III, and the first to make a closure's own internal coherence independently examinable.

FAC: Formal Accountability Collapse

A closure can be internally coherent and still carry no declared, attributable authority behind it. FAC evaluates two declared dimensions - authority and threshold attribution - alongside the already-computed FCC state, resolving one of four structural states: not_detected, detected, critical, or unknown. An absent authority alone is sufficient for a critical state, regardless of the remaining inputs. Not a finding of misconduct, not a determination of legal liability - a narrower structural fact about accountability conditions at closure.

DWC: Decision Wave Compression

The research construct behind this name describes a systemic condition - decision throughput exceeding the capacity of meaningful human oversight, observed across a population of decisions over time. That is not what is implemented today. DWC reads three conditions declared for a single closure - runtime visibility, boundary readiness, unresolved signals - evaluated in a fixed sequence: not_detected, then critical, then detected, then unknown. This article closes Phase III: three mechanisms, three distinct derivations from the same evidentiary profile, none of them a hidden pipeline into the next.

DAPI Identity Binding

Article 07 defined authority conditions through two components: identity and role. DAPI changes the evidentiary status of only one of them. Every EVIDE account requires preventive identity verification through DAPI before activation - but DAPI verifies identity only. It does not validate role, authority, or the correctness of subsequent declarations. Identity is not authority. A verified person is not a verified decision.

GLM-BD

GLM-BD - Governance Layer Manifest, Boundary Declarations - is an open specification. Any governance layer may publish a GLM-compliant manifest declaring what it claims, what it does not claim, where it sits on a timing axis, and what it forbids downstream systems from inferring. EVIDE's own manifest declares itself closure, post-bind, with nine forbidden interpretations and a controlling test: interoperability must increase reviewability without silently transferring authority across layers. Publishing a manifest confers nothing - it makes a self-declared boundary machine-readable, nothing more. This article closes Phase IV.

EAR: Evidentiary Artifact Record

EAR is the first object this series examines that brings together multiple evidentiary states and contextual elements of the same intake into a single versioned record - computed states, declared information, contextual elements, conditional sections, none of it made more true by being preserved together. EAR is versioned and retained, but not a public archive. The current FEDIS workflow freezes a reference to one specific EAR version for a separate certification step, applied outside the EVIDE platform. Preservation is not verification. A record is not a certification.

ANCHOR: Operational Perimeter Declarations

ANCHOR preserves what was explicitly declared about an operational perimeter - its structure, its declarant, its stated authority source, its declared time. A structurally valid Declaration does not authorize an action, does not compare the declared perimeter against what a system actually did, and does not enforce the boundary it describes. A client-supplied declared_at is not, by itself, proof that the Declaration existed at that time. And absence must remain absence: what was not declared is never completed, inferred, or silently strengthened. Preserved is not enforced.

ESB: Epistemic Stabilization Buffer

ESB preserves what was declared while a situation evolved after a decision had already been anchored: observations from a fixed vocabulary, a declared window, and a declared verdict with what it requires (a cause and notes for instability, notes for deferral). The window is declared, not enforced. Observations and verdicts are declared, not verified. What can be reconstructed is the time-referenced sequence of declarations before closure - not the reasoning, the causal process or the actual state of the system behind them. Under normal operation, closure triggers a new EAR version, and it never rewrites the one that showed the window open. A stable verdict is crossing-sufficient, not absolute epistemic truth.