◎ OS PUB Apache 2.0 ← All specifications

P176 — AIEP — Evidence Annotation and Structured Commentary Protocol

Publication Date: 2026-03-27 Status: Open Source Prior Art Disclosure Licence: Apache License 2.0 Author/Organisation: Phatfella Ltd Schema: AIEP_OS_SPEC_TEMPLATE v1.0.1 — https://aiep.dev/schemas/aiep-os-spec-template/v1.0.1

Framework Context

[0001] This disclosure operates within an Architected Instruction and Evidence Protocol (AIEP) environment as defined in United Kingdom patent application number GB2519711.2, filed 20 November 2025, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure defines a protocol for attaching structured, attributed, and versioned AnnotationRecords to evidence artefacts in an AIEP corpus — enabling human reviewers, governance agents, and automated quality-assurance processes to attach contextual commentary, warnings, corrections, endorsements, and caveats to individual artefacts without modifying the canonical artefact content or its ProvenanceChain.

Field of the Disclosure

[0003] This disclosure relates to structured commentary and metadata enrichment protocols for evidence artefacts in governed artificial intelligence systems.

[0004] More particularly, the disclosure concerns: an AnnotationRecord schema; a taxonomy of annotation types; annotation access control; annotation inheritance across artefact versions; annotation surfacing to reasoning chains; and the relationship between annotations and conflict records (P161), quarantine records (P155), and quality metrics (P172).

Background

[0005] Evidence artefacts are canonical content: their text is normalised, their hashes are recorded, and their content must not be modified after ingestion without creating a new version. Yet there is significant value in the ability to attach contextual knowledge to an artefact: a domain expert may identify a methodological caveat in a study; a governance agent may attach a warning about a known conflict (P161) that has not yet been resolved; an automated process may note that an artefact’s content has been superseded in a later publication even if the original artefact remains active.

[0006] An annotation mechanism that is separate from the canonical artefact content allows this contextual enrichment without compromising the integrity guarantees of the artefact itself. Annotations are first-class objects with their own identity, provenance, access control, and versioning.

[0007] Annotations visible to reasoning chains provide a curated, opinionated overlay on the raw evidence corpus — enabling domain experts to communicate contextual knowledge to the reasoning layer without requiring changes to the corpus ingestion pipeline.

Summary of the Disclosure

[0008] AnnotationRecord Schema: Each annotation is represented as:

  • annotation_id — SHA-256 of canonical serialisation of all other fields
  • target_deid — DEID (P162) of the artefact being annotated
  • annotation_type — one of: CAVEAT, ENDORSEMENT, CORRECTION_NOTE, CONFLICT_CROSS_REF, METHODOLOGY_NOTE, SUPERSESSION_NOTE, CONTEXTUAL_BACKGROUND, GOVERNANCE_WARNING
  • annotation_text — structured text body of the annotation (plain text, max 2,048 characters)
  • severity — (for GOVERNANCE_WARNING and CAVEAT types only): INFO, CAUTION, CRITICAL
  • created_by — node fingerprint (P46) or HUMAN_REVIEWER:{reviewer_id} for human-authored annotations
  • created_at — ISO 8601 timestamp
  • reviewed_at — ISO 8601 timestamp of last human review (null if automated)
  • expires_at — optional; annotations may be time-limited (e.g. a GOVERNANCE_WARNING pending conflict resolution has an auto-expiry)
  • linked_records — optional list of references to related records: CONFLICT_RECORD:{conflict_id}, QUARANTINE_RECORD:{quarantine_id}, SUPERSESSION_DEID:{deid}
  • access_level — PUBLIC, GOVERNED_NODE, PLATFORM_ONLY (see [0011])
  • annotation_signature — cryptographic signature by the creating node or reviewer authority
  • superseded_by — annotation_id of the superseding annotation (null if current version)

[0009] Annotation Types: The taxonomy provides semantic precision for reasoning chain consumers:

  • CAVEAT — a methodological, statistical, or contextual caveat that does not invalidate the artefact but should inform interpretation
  • ENDORSEMENT — a positive attestation of artefact quality by a subject-matter authority
  • CORRECTION_NOTE — the artefact contains a known error in a specific field that has been corrected in a later version (links to successor DEID via linked_records)
  • CONFLICT_CROSS_REF — this artefact is a party to an active ConflictRecord (P161); links to CONFLICT_RECORD:{conflict_id}
  • METHODOLOGY_NOTE — contextual information about the evidence-gathering methodology useful for interpreting the artefact’s trust score (P124)
  • SUPERSESSION_NOTE — this artefact has been superseded by a later publication (links to successor DEID)
  • CONTEXTUAL_BACKGROUND — broader contextual knowledge relevant to interpreting the artefact, contributed by an expert reviewer
  • GOVERNANCE_WARNING — governance-layer warning requiring attention (e.g. jurisdiction-specific legal concern, access restriction trigger)

[0010] AnnotationIndex: An AnnotationIndex maintains a mapping from DEID to all active AnnotationRecords for that artefact. Reasoning chains may query the AnnotationIndex for all annotations on a given DEID before processing the artefact, incorporating annotations into their evidence evaluation.

[0011] Annotation Access Control: Annotation visibility is governed by access_level:

  • PUBLIC — visible to all reasoning chains with READ access to the target artefact
  • GOVERNED_NODE — visible only to reasoning chains authenticated via a governance node (P89)
  • PLATFORM_ONLY — visible only to the platform operator; not surfaced to external reasoning chains

[0012] Version Inheritance: When an evidence artefact is updated (new version with new DEID, linked to the predecessor via ProvenanceChain), CAVEAT and METHODOLOGY_NOTE annotations on the predecessor are candidates for inheritance. A AnnotationInheritanceCheck process reviews predecessor annotations and either (a) explicitly copies them to the new DEID with created_by: INHERITED_FROM:{predecessor_deid}; or (b) marks them as NON_APPLICABLE to the new version.

[0013] Automated Annotation Sources: Automated quality processes may create annotations without human review. These annotations include reviewed_at: null and created_by: {process_name}:{node_fingerprint}. Governance nodes (P89) may configure policies requiring human review of automated annotations of GOVERNANCE_WARNING or CONFLICT_CROSS_REF types before they become visible to external reasoning chains.

ASCII Architecture

                  Human Reviewer / Automated Process
                           │  annotation_text, type, severity
                           ▼
                ┌──────────────────────┐
                │  AnnotationRecord    │
                │  (AIEP signed)       │
                └──────────┬───────────┘
                           │
               ┌───────────▼───────────┐
               │    AnnotationIndex    │  ◀── DEID lookup
               │  target_deid → [ann] │
               └───────────┬───────────┘
                           │ query on evidence retrieval
               ┌───────────▼──────────────────────────┐
               │  Reasoning Chain                     │
               │  evaluates annotations alongside     │
               │  artefact content and trust score    │
               └──────────────────────────────────────┘

Annotation Access Control:
  PUBLIC → all chains with READ(deid)
  GOVERNED_NODE → chains via P89 governance node
  PLATFORM_ONLY → operator tooling only

Operational Detail

[0014] Annotation Expiry: Annotations with an expires_at value are marked EXPIRED by the AnnotationIndex at that time; expired annotations are excluded from query results unless the caller explicitly requests historical annotations. Expiry is typically used for GOVERNANCE_WARNING annotations pending resolution of a condition (e.g. embargo lift per P181, conflict resolution per P161).

[0015] Annotation Cardinality Limits: To prevent annotation flooding, the AnnotationIndex enforces a limit on the number of active annotations per DEID per annotation type (default: 5 per type, configurable by governance policy). Submissions exceeding this limit are rejected with a AnnotationLimitExceededError, prompting the submitter to supersede an existing annotation rather than adding a new one.

[0016] Annotation in Evidence Packages: When a reasoning-ready EvidencePackage (P174) is assembled, the PackageAssembler optionally includes all PUBLIC-level annotations for each artefact in the package. The AssemblyRequest (P174 [0011]) includes an include_annotations boolean. When true, each EvidencePackageItem contains an annotations field.

[0017] Quality Metrics Integration: The Evidence Quality Metrics system (P172) incorporates annotation data: the corpus health report includes counts of active GOVERNANCE_WARNING and CAVEAT annotations by severity tier, enabling operators to monitor the proportion of the corpus carrying active warnings.

Claims-Exclusion Notice

This specification is published as open-source prior art. No patent claims are asserted by the author in respect of the mechanisms described. Any third party seeking to patent mechanisms substantially equivalent to those described herein is placed on notice of this prior art disclosure.

This specification is published as open-source prior art under the Apache License 2.0. Copyright 2026 Protocol Owner. Inventor: Protocol Architect.

Patent information · Licensing · All open specifications