◎ OS PUB Apache 2.0 ← All specifications

P160 — AIEP — Evidence Taxonomy and Classification 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 hierarchical taxonomy system for classifying evidence artefacts within AIEP-governed reasoning systems, comprising the AIEP Evidence Taxonomy (a structured classification hierarchy), a TaxonomyClassifier component applied during evidence ingestion, a ClassificationVector attached to each EvidenceNode, and taxonomy-indexed querying of the distributed evidence index (P133) enabling reasoning systems to locate, filter, and weight evidence by subject domain, methodology type, and evidence quality tier.

Field of the Disclosure

[0003] This disclosure relates to evidence taxonomy and classification protocols for governed artificial intelligence reasoning systems.

[0004] More particularly, the disclosure concerns an open hierarchical taxonomy with four top-level dimensions (Subject Domain, Methodology Class, Evidence Quality Tier, and Temporal Class), a ClassificationVector encoding taxonomy assignments for each dimension, a TaxonomyClassifier applying automated multi-label classification during ingestion, a TaxonomyConflict resolution procedure, and query interfaces enabling structured filtering of evidence corpora by taxonomy coordinates.

Background

[0005] AIEP-governed reasoning systems ingest evidence artefacts from heterogeneous sources spanning scientific disciplines, policy domains, technical fields, and grey literature. Without a structured classification system, reasoning queries cannot be scoped to relevant domains, reasoning outputs cannot be weighted by evidence quality tier, and evidence discovery (P133) cannot provide domain-filtered results. An unclassified evidence corpus degrades reasoning precision and makes auditing of evidentiary basis impractical.

[0006] Existing classification systems such as MeSH (medical), JEL (economics), or ACM CCS (computing) are domain-specific and cannot serve as a universal taxonomy for the AIEP general-purpose evidence corpus. A purpose-built AIEP taxonomy is required that covers all subject domains, encodes methodology and evidence quality dimensions relevant to AI-governed reasoning, and is extensible through a governed taxonomy update process.

[0007] The taxonomy must be machine-readable, attached to EvidenceNodes as a structured vector rather than free-text tags, and supported by an automated classifier operating during the normalisation pipeline (P10) to minimise manual annotation burden.

Summary of the Disclosure

[0008] The AIEP Evidence Taxonomy is a four-dimensional hierarchical classification scheme. Each dimension is structured as a tree of codes, where parent codes represent broad categories and leaf codes represent specific sub-domains:

Dimension 1 — Subject Domain (prefix: SD): Root categories include: SD.SCI (Natural Sciences), SD.MED (Medicine & Health), SD.ENG (Engineering & Technology), SD.SOC (Social Sciences), SD.LAW (Law & Regulation), SD.ECO (Economics & Finance), SD.HUM (Humanities), SD.ENV (Environment & Earth), SD.GOV (Government & Policy), SD.MULTI (Multidisciplinary). Each root extends to at least three levels of sub-codes (e.g. SD.MED.ONCO for oncology, SD.MED.ONCO.BREAST for breast cancer).

Dimension 2 — Methodology Class (prefix: MC): MC.EXP (Experimental — randomised or controlled study), MC.OBS (Observational — cohort, case-control, cross-sectional), MC.SYS (Systematic Review or Meta-Analysis), MC.MODEL (Computational or Mathematical Modelling), MC.QUAL (Qualitative Research), MC.THEORY (Theoretical — proof, formal analysis), MC.REV (Narrative Review), MC.STAT (Statistical Dataset), MC.REPORT (Institutional or Government Report), MC.NEWS (News or Journalistic), MC.OTHER (Other or Unclassified).

Dimension 3 — Evidence Quality Tier (prefix: EQ): EQ.T1 (Tier 1 — Peer-reviewed, indexed journal or equivalent), EQ.T2 (Tier 2 — Preprint, working paper, or conference proceedings), EQ.T3 (Tier 3 — Institutional report, government publication, or grey literature), EQ.T4 (Tier 4 — Journalistic, editorial, or opinion), EQ.T5 (Tier 5 — Unverified, user-generated, or unknown quality).

Dimension 4 — Temporal Class (prefix: TC): TC.CURR (Current — published within 2 years), TC.RECENT (Recent — 2–10 years), TC.HIST (Historical — more than 10 years), TC.PERM (Permanent — canonical reference unlikely to be superseded, e.g. mathematical proofs, legal statutes), TC.LIVE (Live — continuously updated).

[0009] ClassificationVector: The ClassificationVector attached to each EvidenceNode is a structured object:

{
  "subject_domain": ["SD.MED.ONCO.BREAST", "SD.SCI.BIO.GEN"],
  "methodology_class": "MC.EXP",
  "evidence_quality_tier": "EQ.T1",
  "temporal_class": "TC.CURR",
  "classifier_version": "AIEP_TAX_v1.3",
  "classification_confidence": 0.91,
  "classification_source": "AUTO"
}

subject_domain is multi-valued (up to eight codes). All other dimensions are single-valued. classification_source is AUTO (automated classifier), MANUAL (human annotator), or HYBRID (automated with human review).

[0010] TaxonomyClassifier: The TaxonomyClassifier is a multi-label classifier applied as a step in the normalisation pipeline (P10). It accepts the EvidenceNode’s content_canonical text and citation metadata (P157) as inputs and produces a ClassificationVector. The classifier assigns Subject Domain codes by semantic analysis of the text, assigns Methodology Class from structured signals in the methodology section or abstract, assigns Evidence Quality Tier from publisher type and peer-review status signals in the citation metadata, and assigns Temporal Class from the year and citation_type fields of the CitationRecord.

[0011] TaxonomyConflict Resolution: Where the automated classifier assigns a code with confidence below a configurable threshold (default: 0.70), the EvidenceNode is flagged as CLASSIFICATION_UNCERTAIN and queued for human review. Where human review produces a different code from the automated assignment, the human assignment overrides and the classification_source is updated to MANUAL. Both the automated and human assignments are retained in the ledger.

[0012] Taxonomy Querying: The distributed evidence index (P133) provides taxonomy-filtered query endpoints. A TaxonomyQuery specifies one or more ClassificationVector dimension values as filters; the index returns all EvidenceNodes matching all specified filters. Queries may use partial codes to retrieve all descendants of a parent (e.g. SD.MED matches all medical evidence). Multiple Subject Domain codes in a query are combined with OR semantics; all other dimension filters use AND semantics.

[0013] Taxonomy Governance: The AIEP Evidence Taxonomy is versioned under semantic versioning. New codes may be added in minor versions. Codes may be deprecated but not removed in minor versions (deprecated codes are retained for historical artefacts). Breaking restructuring requires a major version increment. Taxonomy updates are published by the AIEP taxonomy governance committee and announced via the well-known discovery mechanism (P64). Nodes update their TaxonomyClassifier and their IndexFilter configuration on announcement receipt.

ASCII Architecture

EvidenceNode (raw)
        │
        ▼
┌─────────────────────┐
│  TaxonomyClassifier │◀── AIEP Evidence Taxonomy v1.x
│  - Subject Domain   │    (SD / MC / EQ / TC codes)
│  - Methodology      │
│  - Quality Tier     │
│  - Temporal Class   │
└──────────┬──────────┘
           │ ClassificationVector
           ▼
┌─────────────────────┐     ┌────────────────────────┐
│  EvidenceNode       │────▶│  Evidence Index (P133) │
│  .classification =  │     │  Taxonomy-indexed       │
│    ClassVector      │     └────────────────────────┘
└─────────────────────┘
           │ uncertain flag
           ▼
┌─────────────────────┐
│  Human Review Queue │──▶ MANUAL override
└─────────────────────┘

Operational Detail

[0014] Downstream Weight Adjustment: The Evidence Quality Tier code in the ClassificationVector is used by the AIEP trust scoring system (P124) as one input to the trust weight assigned to an artefact. An EQ.T1 artefact from a peer-reviewed source receives a higher base trust weight than an EQ.T4 editorial. This interaction is defined in P124; P160 defines only the taxonomy and classification procedure.

[0015] Federation Integration: Federated evidence providers (P156) publish the Subject Domain taxonomy codes (subject_scope) in their FederationContracts using the AIEP Evidence Taxonomy SD-prefix codes. This aligns federation scoping with the internal taxonomy, enabling StreamBroker subscriptions (P158) to filter streaming events by taxonomy code.

[0016] Extensibility: Organisations deploying AIEP may register private taxonomy subtrees under the SD.PRIV.* prefix for domain-specific classification codes not covered by the standard taxonomy. Private codes are not interoperable across organisations but follow the same ClassificationVector format and are transparently handled by the query and indexing infrastructure.

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.