◎ OS PUB Apache 2.0 ← All specifications

P184 — AIEP — Evidence Corroboration Score 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 computing and maintaining a CorroborationScore for each evidence artefact — a normalised measure of the degree to which independent sources in the corpus corroborate the primary claim or finding of that artefact — distinct from the trust score (P124, which measures source quality) and the freshness score (P147, which measures temporal currency).

Field of the Disclosure

[0003] This disclosure relates to multi-source corroboration measurement protocols for evidence artefacts in governed artificial intelligence reasoning systems.

[0004] More particularly, the disclosure concerns: a CorroborationScore computation algorithm; corroborating artefact identification; independence verification to prevent circular corroboration; corroboration score update triggers; the integration of corroboration scoring with trust scoring (P124), conflict detection (P161), and reasoning chain evidence evaluation; and the distinction between corroboration and replication in the cross-reference network (P177).

Background

[0005] Trust scoring (P124) provides a measure of the credibility of an evidence source — its publication venue, author authority, peer review status, and similar factors. A high trust score indicates the source is credible by its nature; it does not indicate that other independent sources agree with the claim made.

[0006] Corroboration scoring addresses this orthogonal dimension: how many independent sources in the corpus make the same claim, finding, or measurement? A finding from a single high-trust source that no other source corroborates is more epistemically uncertain than the same finding corroborated by five independent high-trust sources.

[0007] Independence verification is crucial: if Source B simply cites Source A without independent evidence, counting Source B as a corroborator of Source A would be circular. True corroboration requires that the corroborating artefact’s claim is based on independently collected evidence — different methodology, different investigator, different dataset.

Summary of the Disclosure

[0008] CorroborationScore Definition: The CorroborationScore for an artefact A is a value in [0.0, 1.0] computed as: corroboration_score(A) = f(corroborating_count, corroborating_trust_scores, independence_weights)

where corroborating_count is the number of artefacts determined to independently corroborate A’s primary claim; corroborating_trust_scores is the set of trust scores (P124) of the corroborating artefacts; and independence_weights encode the degree of independence of each corroborating source from A.

[0009] Corroborating Artefact Identification: An artefact B corroborates artefact A if:

B’s ClassificationVector (P160) places it within the same taxonomy leaf node or parent node as A
B makes a claim or finding that is semantically congruent with A’s primary claim (verified by semantic analysis producing a congruence_score ≥ 0.7)
B’s CrossReferenceEdge (P177) to A (if any) is of type REPLICATES or CITES_AS_SUPPORTING (not CONTRADICTS, FAILS_TO_REPLICATE, or CORRECTS)
B was not produced by the same research group or institution as A (independence condition; see [0010])

[0010] Independence Verification: The independence_weight of corroborator B relative to primary A is:

1.0 — B has no author overlap with A (checked via CitationRecord author lists, P157), and B was published by a different institution
0.7 — B was published by a different institution but shares one or more non-lead authors with A
0.4 — B is from the same institution but different research group (different lead author)
0.0 — B is from the same research group as A (same lead author, same institution) — excluded from corroboration scoring as non-independent

[0011] CorroborationScore Computation: Given N qualified corroborators: raw_score = Σ (trust_score(Bᵢ) × independence_weight(Bᵢ)) for i = 1..N corroboration_score = min(1.0, raw_score / corroboration_saturation_threshold)

where corroboration_saturation_threshold is configurable (default: 5.0 — representing five fully independent, high-trust corroborators as the maximum corroboration achievable). A single perfect corroborator (trust_score = 1.0, independence_weight = 1.0) yields corroboration_score = 0.2; five such corroborators yield corroboration_score = 1.0.

[0012] CorroborationRecord Schema: The computed score and its derivation are stored as a CorroborationRecord:

corroboration_record_id — SHA-256 of canonical serialisation
subject_deid — DEID of the artefact being scored
corroboration_score — value in [0.0, 1.0]
corroborating_deids — list of DEIDs of the qualifying corroborating artefacts with their trust_score, independence_weight, and congruence_score
disqualified_deids — list of DEIDs considered but disqualified (with disqualification reason): INSUFFICIENT_CONGRUENCE, NON_INDEPENDENT_SOURCE, CONTRADICTS_TARGET, FAILS_TO_REPLICATE
computed_at — ISO 8601 timestamp
computed_by — node fingerprint (P46)
record_signature — cryptographic signature

[0013] Corroboration Score Update Triggers: The CorroborationScore is recomputed when:

A new artefact is admitted to the corpus in the same taxonomy domain as the subject artefact (potential new corroborator)
The trust score (P124) of an existing corroborating artefact changes
A new CrossReferenceEdge (P177) is asserted between the subject artefact and another artefact
A ConflictRecord (P161) is opened or resolved involving the subject artefact or a corroborating artefact

[0014] Conflict Impact on Corroboration: If a qualifier corroborator B has an open ConflictRecord (P161) with artefact C where C CONTRADICTS B with conflict_status: OPEN, the congruence_score contribution of B to A’s CorroborationScore is weighted down by (1 - conflict_severity_weight) until the conflict is resolved.

ASCII Architecture

Subject Artefact A
        │
        ▼ identify candidates
Artefacts in same taxonomy domain
        │
        ▼ semantic congruence filter (≥ 0.7)
Congruent Artefacts
        │
        ▼ independence check (author/institution)
Independence Weights (1.0, 0.7, 0.4, 0.0)
        │ disqualify non-independent (0.0)
        ▼
Qualified Corroborators: B₁..Bₙ
        │
        ▼ compute
raw_score = Σ trust_score(Bᵢ) × independence_weight(Bᵢ)
corroboration_score = min(1.0, raw_score / saturation_threshold)
        │
        ▼
CorroborationRecord (stored per DEID)
        │
        ├──▶ Trust scoring (P124): complement to trust_score
        ├──▶ Conflict detection (P161): open conflicts reduce weight
        ├──▶ Evidence packages (P174): corroboration_score in item
        └──▶ Quality metrics (P172): corpus corroboration distribution

Operational Detail

[0015] Corroboration Score in Evidence Packages: The EvidencePackageItem (P174) includes corroboration_score as a field. Reasoning chains receiving a package can thus evaluate both source quality (trust_score) and independent support (corroboration_score) for each artefact without separate lookups.

[0016] Corroboration Visibility to Reasoning Chains: The corroboration score and its derivation (corroborating DEIDs) are surfaced to reasoning chains on request, enabling them to retrieve and directly evaluate the corroborating artefacts rather than accepting the score as a black-box metric. The corroborating_deids list in the CorroborationRecord provides the entry points for this secondary retrieval.

[0017] Distinguishing Corroboration from Replication: The cross-reference edge type REPLICATES (P177) is a specific relationship indicating a formal replication study. CorroborationScore is a broader measure that includes any independently produced artefact that reaches the same conclusion, regardless of whether it was formally designated as a replication study. Formal replications (CrossReferenceEdge: REPLICATES) automatically qualify as corroborators if they pass the independence and congruence filters.

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.