Glossary

A reference glossary for terms used across the AIEP protocol and this Hub.

Admissibility Gate

The conjunction of the Plausibility Gate and Probability Certification check that governs whether an artefact may be used for execution. Corresponds to Layer 3b of the AIEP stack. An artefact can be structurally valid, hash-verified, and issuer-resolved and still fail the admissibility gate. See /plausibility and /probability-engine.

AIEP

AIEP stands for Architected Instruction & Evidence Protocol. It is a publishing and retrieval protocol that lets organisations publish structured, verifiable artefacts in a form that both humans and AI systems can retrieve and validate.

Artefact

A structured, schema-conformant JSON document published by an AIEP publisher. Artefacts are the atomic unit of AIEP - everything published under the protocol is an artefact.

Artefact hash

A SHA-256 hash of the canonical content of an artefact, recorded in the Mirror index at publication. A retriever can recompute the hash and compare it to the index entry to confirm the artefact has not been altered.

CanonicalHash

The SHA-256 hash computed over the canonical serialisation of an artefact. The canonical serialisation is fully deterministic: fields are ordered, whitespace is normalised, and encoding is UTF-8. Different nodes computing the CanonicalHash of the same artefact must produce bit-identical results. Any divergence indicates either tampering or a non-canonical serialisation path.

Canonical schema

The authoritative JSON Schema definition for a given artefact type. An artefact is structurally valid if and only if it conforms to its canonical schema. Schemas are versioned and published at /.well-known/aiep/schemas/.

Certificate

An AIEP_CERTIFICATE artefact issued by a registered AIEP authority, linking a publisher to a certified claim. Certificates carry an issuer identity, a validity period, and a hash.

Certification

An optional claim that a system meets a defined standard and can prove it. In AIEP, certification exists solely to protect the meaning of “AIEP Certified”. It does not restrict protocol adoption.

Compliance

The process of verifying certification claims and detecting redistribution of NDA-restricted materials. Compliance does not police open adoption; it protects trust when certification is claimed.

Compliance signals

Machine-readable indicators of a publisher’s compliance status, published at /.well-known/aiep/compliance/signals.json. Typed rules (schema conformance, hash integrity, registry-linked) with pass/fail status and evidence references.

ComplianceCertificate

An AIEP_COMPLIANCE_CERTIFICATE artefact asserting that a specific output meets a defined regulatory or operational standard. Bound to a named output artefact, an evidence chain, and an issuing authority. Required by enterprise deployments operating under audit regimes. Not the same as a publisher Certificate.

Conjunction trust model

AIEP’s trust architecture, which differs fundamentally from PKI certificate hierarchies. Trust requires all four conditions to hold simultaneously: Identity (issuer resolves in registry), Integrity (hash verified), Admissibility (plausibility and probability gates pass), Goal commitment (GoalVector committed and non-drifted). Failure of any one condition produces a non-trusted artefact regardless of the others. See /security.

ContextReconstructionHash

The hash computed over a fully reconstructed historical context state by the P22 recall engine. Because reconstruction is deterministic - using the same canonical primitives over the same archived artefacts - all distributed nodes produce bit-identical ContextReconstructionHash values. A mismatch indicates incomplete or corrupted archives and triggers fail-closed behaviour.

Dissent fork

A branch created when a divergent interpretation or evidence chain encounters a plausibility or probability gate failure. The branch is not discarded - it is archived with a complete DivergenceRecord and a typed PlausibilityDeclarationHash or ProbabilityEnvelopeHash. The dissent archive is the source of recall candidates when registry conditions change.

DivergenceGraph

The directed acyclic graph representing all evidential, temporal, interpretive, or branch divergences detected during a reasoning operation. Each node carries evidence, a DivergenceType, a confidence bound, and a GoalVector reference. The DivergenceGraph is the primary input to both the probability certification engine and the plausibility gate evaluation. See /divergence.

Evidence

The supporting artefact or set of artefacts that justify an instruction. Evidence can be a document, a dataset, a log, a certificate, a measurement, or any other object that can be referenced and validated.

EvidenceWeightVector

The structured vector assigning a weight to each individual evidence artefact contributing to a PlausibilityScore or canonical scoring function output. Weights are registry-bound, deterministically aggregated, and included in the DivergenceRecord so the scoring can be independently reproduced.

Fail-closed

If required proof is missing, a claim is treated as invalid. In AIEP this applies to certification claims, admissibility checks (if plausibility or probability gate data is absent), and recall operations (if archive artefacts are missing). There is no fail-open fallback in the protocol core.

GENOME SDK

The open-source developer library for building AIEP-conformant hardware execution substrates. The SDK enforces the four GENOME constitutional invariants at the kernel level. Source-available under Apache 2.0 licence. See /genome-sdk.

GENOME_LOCKFILE

The versioned manifest that pins a deployment to a specific GENOME kernel version and canonical primitive set. Written to the deployment manifest at build time and verified at runtime. Post-deployment kernel substitution is architecturally blocked by the lockfile mechanism.

GoalVector

A structured, cryptographically committed record of an AI system’s goal state at a specific point in time. Carries a hash, a temporal stamp, a GoalArtefact reference, and a drift detection record. Execution operations must carry a non-drifted GoalVector reference to pass the conjunction trust model’s goal commitment condition. See /goal-generation.

Innovation Ledger

An append-only log of timestamped innovation records. Each entry is an AIEP_INNOVATION_LEDGER_ENTRY artefact establishing provenance for a piece of intellectual work.

Instruction

Any claim, directive, decision, or statement that is intended to be relied upon. The key question for every instruction is: what evidence supports it?

Issuer

A registered AIEP authority permitted to issue certificates. Issuers are listed in the AIEP registry with their public identity and scope of certification.

Layer 3b

The admissibility gate layer of the AIEP stack, sitting between Layer 3 (Evidence Ecosystem) and Layer 4 (Constitutional Stack). Comprises the Plausibility Matrix (P03), Probability Certification Engine (P04), Quantum Alignment Layer (P02), and the P66/P67 public declaration formats. Evidence that passes Layer 3 is not automatically execution-eligible until it also passes Layer 3b. See /stack-integration.

Mirror

The structured, machine-readable surface hosted by an AIEP publisher at /.well-known/aiep/. A Mirror contains an index, metadata, and all published artefacts. It is the primary discovery and retrieval endpoint for AI systems.

NormalisationProfile

A registry-bound set of normalisation rules defining how artefacts of a given type are canonicalised before hashing and scoring. The NormalisationProfile is versioned independently. All nodes applying the same NormalisationProfile produce bit-identical canonical representations. Included in the NormalisationManifest artefact.

PlausibilityDeclarationHash

The hash computed over a P67-format PlausibilityDeclaration document attached to an artefact. The P67 format is publicly released under Apache 2.0 and defines the binding between a claim-type, its PlausibilityScore, the registry version that produced it, and the Merkle proof of inclusion. Third parties can independently verify the declaration without AIEP protocol access.

PlausibilityScore

The numeric score assigned to a claim-type by an authorised registry authority, reflecting the evidence-weighted probability of the claim-type’s admissibility under current knowledge. Bounded [0, 1]. Retrieval, Merkle proof verification, and threshold comparison against the two execution bands are performed by the Plausibility Matrix gate (P03). See /plausibility.

ProbabilityEnvelopeHash

The hash computed over a P66-format ProbabilityEnvelope document. P66 is publicly released under Apache 2.0 and defines the binding between a DivergenceGraph node, its certified tail-risk bound epsilon, the confidence interval C, the computation method, and the cryptographic commitment. Enables third-party verification of probability certification without privileged access.

Publisher

An organisation or individual that hosts an AIEP Mirror and publishes artefacts under the protocol. Publishers may be registered (listed in the AIEP registry) or unregistered (self-hosted, open use).

Publisher identity

A stable identifier for a publisher - typically their domain (e.g. example.com) - declared in metadata.json and used across all artefacts. Registered publishers are resolvable via the AIEP registry.

Quantum Alignment Layer (QAL)

The protocol layer (P02 / GB2519798.9) that ensures quantum and classical computation paths produce deterministically equivalent results before any commitment is made. Runs quantum and classical scoring in parallel, canonicalises both, computes a deterministic deviation metric. If equivalence is confirmed, the result is committed. If not, the QAL fails closed and the classical result is used as the sole committed value. See /quantum-alignment.

Record type

The record_type field on every AIEP artefact. Declares which canonical schema the artefact conforms to.

RecallScope

The bounded set of archived artefacts and divergence records that constitute a valid recall operation. Defined by P83 to prevent unbounded historical re-evaluation. A RecallScope specifies the archived branch, the plausibility threshold change that triggered eligibility, the time bounds, and the authorised recall initiator. Reconstruction outside a valid RecallScope is not a protocol-compliant recall.

Registry

The AIEP registry of registered publishers and issuers. Accessible at /.well-known/aiep/registry/index.json. Used by retrievers to resolve publisher identity and confirm issuer authority.

Schema

See Canonical schema.

Tail-risk bound

The certified upper bound on the failure probability of a DivergenceGraph node: expressed as Failure probability <= epsilon at confidence C. For safety-critical AIEP operations epsilon <= 10^-15. Computed by the Probability Certification Engine (P04). Fail-closed if the bound cannot be established within the required tolerance or if independent recomputation produces a non-identical result.

Validator

A tool or service that checks an artefact against its canonical schema and verifies its hash. The AIEP Hub includes a built-in validator at /validator.

Visibility

Describes whether an innovation ledger entry is fully public or has NDA-gated detail. In an NDA entry, the public record still publishes a sealed hash to prove existence without disclosure.

Well-known path

The conventional base path for AIEP Mirror content: /.well-known/aiep/. Follows the IETF well-known URI convention (RFC 8615). The presence of index.json at this path signals that the host is an AIEP publisher.

See: Protocol · Schema Catalogue · Mirror · Security