◎ OS PUB Apache 2.0 ← All specifications

P170 — AIEP — Evidence Dependency Graph 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 building, maintaining, and querying a directed Evidence Dependency Graph (EDG) within AIEP-governed evidence infrastructure, representing dependency relationships between evidence artefacts such as meta-analyses depending on constituent studies, systematic reviews depending on included papers, and inferred artefacts (P185) depending on their source artefacts, enabling cascade impact analysis when any node in the graph is updated or invalidated.

Field of the Disclosure

[0003] This disclosure relates to evidence dependency graph protocols for governed artificial intelligence reasoning systems.

[0004] More particularly, the disclosure concerns an EvidenceDependencyLink schema encoding typed directional dependencies between evidence artefacts; a DependencyGraph maintained as a partition of the Deterministic Dual-Ledger Memory Substrate (P80); a DependencyGraphQuerier enabling forward (what artefacts depend on this one?) and reverse (what does this artefact depend on?) traversal; and a CascadeImpactAnalyser computing the set of artefacts affected when a given artefact is updated, invalidated, or retracted.

Background

[0005] Evidence artefacts in AIEP corpora are not independent: a meta-analysis depends on the individual studies it synthesises; a systematic review depends on the papers it includes; a news report depends on the government statistics it cites; an inferred artefact (P185) depends on the source artefacts from which it was inferred. When an upstream artefact is updated or retracted, all downstream artefacts that depend on it may need to be re-evaluated or invalidated.

[0006] The Evidence Provenance Chain (P150) records the transformation history of individual artefacts but does not model cross-artefact dependencies. Cross-references in CitationRecords (P157) record bibliographic attribution but not the semantic dependency type. A dedicated dependency graph is required to make these relationships explicit, queryable, and usable for cascade impact analysis.

[0007] Dependency graphs in evidence corpora have well-defined practical uses: when a foundational study is retracted, reasoning systems must identify all downstream artefacts (meta-analyses, reviews, inferences) that relied on it and re-evaluate their validity. Without a queryable dependency graph, this cascade analysis cannot be performed systematically.

Summary of the Disclosure

[0008] EvidenceDependencyLink Schema: A directed, typed link between two evidence artefacts:

  • link_id — SHA-256 of canonical serialisation of all other fields
  • source_deid — DEID of the upstream (dependency) artefact
  • target_deid — DEID of the downstream (dependent) artefact
  • dependency_type — enumeration:
    • CONSTITUENT — target includes source as a constituent study or dataset (e.g. meta-analysis → primary study)
    • CITES — target formally cites source in its reference list
    • REPLICATES — target is a replication study of source
    • REFUTES — target provides evidence that refutes source
    • SUPPORTS — target provides corroborating evidence for source
    • INFERRED_FROM — target was inferred from source by a reasoning operation (P185)
    • UPDATES — target is an updated version of source (different from temporal supersession, P161; used when target is explicitly issued as an update by the same originating body)
    • TRANSLATES — target is a translated version of source (beyond P159; used for independently produced translations)
  • dependency_confidence — float 0–1 (confidence that the dependency exists; 1.0 for explicit links from structured metadata; lower for automatically extracted links)
  • link_source — STRUCTURED (from CitationRecord or explicit metadata), EXTRACTED (automatically extracted from artefact text), INFERRED (from reasoning operations P185), MANUAL (human curator)
  • created_at — ISO 8601 timestamp

[0009] Evidence Dependency Graph (EDG): The EDG is a directed graph stored in the DEPENDENCY_GRAPH partition of the Deterministic Dual-Ledger Memory Substrate (P80). Nodes are DEIDs; edges are EvidenceDependencyLinks. The graph is append-only (links are added, never deleted; superseded links are marked DEPRECATED). The EDG is accessible to all AIEP nodes with READ access to the ledger partition.

[0010] DependencyGraphQuerier: The DependencyGraphQuerier provides two traversal operations:

  • forward_dependencies(deid, depth, types) — returns all artefacts that the given DEID depends on, up to depth hops, optionally filtered by dependency_type
  • reverse_dependencies(deid, depth, types) — returns all artefacts that depend on the given DEID, up to depth hops

Results include the full EvidenceDependencyLink for each edge and the ClassificationVector (P160) and CitationRecord (P157) of each node.

[0011] Automatic Link Extraction: The DependencyLinkExtractor applies during evidence ingestion to automatically identify dependency links from artefact content. It extracts: (a) CITES links from the reference list in the CitationRecord and from DOI cross-references; (b) CONSTITUENT links from structured metadata in systematic review and meta-analysis artefacts declaring included studies; (c) REPLICATES and REFUTES links from explicit language in the artefact abstract (keyword matching). All automatically extracted links are assigned link_source = EXTRACTED and dependency_confidence below the explicit structured metadata confidence floor (default: 0.80).

[0012] CascadeImpactAnalyser: When an artefact is updated (new ProvenanceLink added), invalidated, or retracted (P161), the CascadeImpactAnalyser performs a reverse dependency traversal from the affected DEID and produces a CascadeImpactReport:

  • trigger_deid — the DEID that changed
  • trigger_event — UPDATE, INVALIDATION, RETRACTION
  • affected_artefacts — ordered list by traversal depth of all DEIDs reachable via reverse dependencies
  • highest_dependency_type — the most significant dependency type in the impact chain
  • recommended_action — RE_EVALUATE (artefact should be re-evaluated given the change), QUARANTINE (artefact should be quarantined pending review, P155), or MONITOR (low-confidence link; monitor but no immediate action)

[0013] INFERRED_FROM Links: When the Evidence Synthesis and Inference Protocol (P185) produces an inferred artefact, it automatically adds INFERRED_FROM links from each source artefact to the inferred artefact. This ensures the EDG captures the full provenance of inferred artefacts and any retraction or update to a source artefact triggers a CascadeImpactReport for all artefacts inferred from it.

ASCII Architecture

EvidenceArtefact A (meta-analysis)
  │ CONSTITUENT (explicit)         CONSTITUENT (explicit)
  │─────────────────────────────────────────────────┐
  ▼                                                 ▼
Study B                                          Study C
  │ CITES (extracted)
  ▼
Background Paper D

Retraction of Study B:
  CascadeImpactAnalyser
  reverse_dependencies(Study_B) →
      A: CONSTITUENT (RE_EVALUATE)
      All artefacts INFERRED_FROM A: QUARANTINE

EDG partition (P80):
[link: B→A, CONSTITUENT, confidence=1.0]
[link: B→D, CITES, confidence=0.85]

Operational Detail

[0014] Cycle Detection: The DependencyGraphQuerier detects cycles (artefact A → B → A) during traversal and breaks cycles on re-visit, reporting any cycles found as DependencyCycleRecords in the ledger. Cycles in a citation graph indicate a data quality issue (artefacts incorrectly citing each other) and are flagged for human review.

[0015] Confidence Calibration: The dependency_confidence score for automatically extracted links is calibrated against a curated benchmark set of known-correct and known-incorrect dependency links. Calibration is performed when the DependencyLinkExtractor model is updated and the results are stored in the node’s model registry.

[0016] Graph Partitioning: For very large corpora, the EDG is partitioned across nodes using the same consistent hashing scheme as the DEID Registry (P162). Each partition covers a range of source DEIDs. Cross-partition traversal queries are routed through the hub network.

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