P112 — AIEP — Atomic Claim Extraction and Cross-Verification Layer
applicant: Neil Grassby inventor: Neil Grassby status: review — to file classification: Patent Application — Confidential priority: Claims priority from GB2519711.2 filed 20 November 2025
Field of the Invention
[0001] The present invention relates to evidence admission quality control in governed reasoning substrates.
[0002] More particularly, the invention relates to a mechanism within an AIEP corpus admission pipeline for decomposing an admitted evidence artefact into its constituent atomic factual claims, cross-verifying each claim against existing Evidence Ledger contents, generating ContradictionRecords where new claims contradict existing high-weight evidence, and applying a ConditionalWeightAdjustment to the new artefact’s evidence weight pending contradiction resolution.
Background
[0003] The AIEP corpus admission pipeline (P10, P17, P24, P25) normalises admitted documents and computes evidence weight as a function of provenance, corroboration count, recency, and schema validity. These properties evaluate the document as a unit. They do not evaluate the correctness of individual factual claims within the document.
[0004] A document may carry high provenance and corroboration scores while containing specific factual claims that contradict well-established high-weight evidence already in the substrate. Under prior admission mechanisms, such a document enters the substrate at high evidence weight and immediately influences active reasoning branches, without any detection of the internal contradiction.
[0005] The problem is particularly acute for scientific literature, where retracted or contested papers may have accumulated high corroboration scores before their invalidation, and for regulatory documents, where jurisdiction-specific provisions may directly contradict provisions in other jurisdictions already admitted to the substrate.
[0006] Existing systems do not provide: decomposition of admitted artefacts into atomic factual claims prior to Evidence Ledger commitment; cross-verification of each atomic claim against existing substrate evidence; ContradictionRecord generation binding the contradicting claims and their source artefacts; or conditional evidence weight adjustment pending contradiction resolution.
Summary of the Invention
[0007] A ClaimExtractionOperation is executed as a step in the corpus admission pipeline after P25 cross-domain stitching and before P94 monitor check.
[0008] The ClaimExtractionOperation applies a schema-defined ClaimExtractionProfile to the admitted artefact’s CanonicalForm, decomposing it into an ordered set of AtomicClaims. Each AtomicClaim comprises:
(a) ClaimId — deterministic identifier computed as H(ClaimText ‖ ArtefactHash ‖ ClaimPosition ‖ SchemaVersionId); (b) ClaimText — canonical natural language representation of the claim; (c) ClaimDomain — domain classification per P24; and (d) ClaimConfidenceScore — a normalised score [0.0, 1.0] representing the structural confidence of the extraction.
[0009] A ClaimCrossVerificationOperation evaluates each AtomicClaim against existing Evidence Ledger contents. For each existing high-weight evidence artefact in the same ClaimDomain, a ContradictionScore is computed as a deterministic function of the semantic opposition between the new AtomicClaim and the existing artefact’s claim set.
[0010] Where ContradictionScore exceeds a schema-defined ContradictionThreshold for any existing artefact, a ContradictionRecord is generated and appended to the Evidence Ledger comprising:
(a) NewArtefactHash; (b) ConflictingArtefactHash; (c) ConflictingClaimId; (d) ContradictionScore; (e) ContradictionHash = H(NewArtefactHash ‖ ConflictingArtefactHash ‖ ConflictingClaimId ‖ ContradictionScore ‖ SchemaVersionId); and (f) timestamp.
[0011] Upon ContradictionRecord generation, a ConditionalWeightAdjustment is applied to the new artefact’s evidence weight:
AdjustedEvidenceWeight = BaseEvidenceWeight × (1 - ContradicitionDiscountFactor × NormalisedContradictionScore)
where ContradicitionDiscountFactor is a schema-defined version-bound parameter.
[0012] The ConditionalWeightAdjustment is reversible. Upon ContradictionRecord resolution — when subsequent evidence confirms either the new artefact or the conflicting artefact — the weight adjustment is updated accordingly and a ContradictionResolutionRecord is appended.
[0013] Existing Evidence Ledger artefacts that are contradicted by the new artefact’s AtomicClaims receive a corresponding ContradictionRecord appended to their ledger entries, with ConditionalWeightAdjustment applied symmetrically.
[0014] A ClaimExtractionManifest is generated for each admitted artefact comprising: all extracted AtomicClaimIds; ClaimExtractionProfile version; ContradictionRecords generated; and ClaimManifestHash = H(CanonicalSerialise(AtomicClaimIds) ‖ ClaimExtractionProfileVersion ‖ ArtefactHash ‖ SchemaVersionId).
Claims
-
An atomic claim extraction and cross-verification system within an AIEP corpus admission pipeline, the system configured to: decompose an admitted evidence artefact into AtomicClaims using a schema-defined ClaimExtractionProfile applied to the artefact’s CanonicalForm; evaluate each AtomicClaim against existing Evidence Ledger contents by computing a ContradictionScore as a deterministic function of semantic opposition; generate a ContradictionRecord upon ContradictionScore exceeding a schema-defined threshold, binding the new artefact, conflicting artefact, and contradiction score by cryptographic hash; apply a ConditionalWeightAdjustment to the new artefact’s evidence weight as a function of the ContradictionScore; and append a ClaimExtractionManifest to the Evidence Ledger binding all extracted AtomicClaimIds and ContradictionRecords.
-
The system of claim 1 wherein ConditionalWeightAdjustment is computed as BaseEvidenceWeight multiplied by a discount factor that is a deterministic function of the normalised ContradictionScore.
-
The system of claim 1 wherein ConditionalWeightAdjustment is reversible upon ContradictionRecord resolution, with a ContradictionResolutionRecord appended to the immutable Evidence Ledger.
-
The system of claim 1 wherein ContradictionRecords are appended to both the new artefact’s ledger entry and the conflicting artefact’s ledger entry symmetrically.
-
The system of claim 1 wherein ClaimExtractionManifest is cryptographically bound to the artefact hash and ClaimExtractionProfile version, enabling deterministic replay of the extraction operation.
-
The system of claim 1 wherein ClaimExtractionOperation executes after P25 cross-domain stitching and before P94 monitor check in the corpus admission pipeline, such that contradiction detection precedes branch surfacing evaluation.
-
A method for atomic claim cross-verification in an AIEP corpus admission pipeline comprising: extracting AtomicClaims from an admitted artefact’s CanonicalForm; computing ContradictionScores against existing Evidence Ledger contents; generating ContradictionRecords upon threshold crossing; applying ConditionalWeightAdjustment; and generating a ClaimExtractionManifest.
-
A computing system comprising one or more processors and memory storing instructions which, when executed, perform the method of claim 7.
Drawings
FIG. 1 — Architecture diagram (see filed application for figures)
Figure 1 — ClaimExtractionOperation Pipeline
Admitted Evidence Artefact (CanonicalForm)
│
▼
ClaimExtractionOperation
uses ClaimExtractionProfile (schema-defined)
│
▼
AtomicClaims[] extracted
deterministically from CanonicalForm
│
▼
ClaimCrossVerificationOperation
evaluates each AtomicClaim vs
existing Evidence Ledger contents
Figure 2 — ContradictionScore and ConditionalWeightAdjustment
AtomicClaim[i] vs Evidence Ledger entry[j]
│
▼
ContradictionScore[i,j] = f(semantic_divergence,
domain_compatibility)
│
├── ContradictionScore < threshold ──► no action
└── ContradictionScore >= threshold ──►
ContradictionRecord generated
(binds both artefacts by hash)
ConditionalWeightAdjustment applied
to new artefact's evidence weight
(adjustment is reversible on resolution)
Figure 3 — Symmetric ContradictionRecord Ledger Append
ContradictionRecord
new_artefact_hash
conflicting_artefact_hash
ContradictionScore
ConditionalWeightAdjustment value
│
▼
Appended to new artefact's ledger entry
Appended to conflicting artefact's ledger entry
(symmetric — both entries reference each other)
Figure 4 — ClaimExtractionManifest and Pipeline Position
┌────────────────────────────────────┐
│ ClaimExtractionManifest │
│ artefact_hash │
│ AtomicClaims[] │
│ ContradictionRecords[] │
│ ClaimExtractionProfile version │
│ ManifestHash = H(all fields) │
└──────────────────┬─────────────────┘
│
Pipeline position:
P25 cross-domain stitching
→ ClaimExtractionOperation <- here
→ P94 monitor evaluation
Abstract
An atomic claim extraction and cross-verification system is disclosed for AIEP corpus admission pipelines. Upon evidence artefact admission, a ClaimExtractionOperation decomposes the artefact’s CanonicalForm into AtomicClaims using a schema-defined ClaimExtractionProfile. A ClaimCrossVerificationOperation evaluates each AtomicClaim against existing Evidence Ledger contents, computing a ContradictionScore for each comparison. Where ContradictionScore exceeds a threshold, a ContradictionRecord is generated binding both artefacts by cryptographic hash and a ConditionalWeightAdjustment is applied to the new artefact’s evidence weight. The adjustment is reversible upon resolution. ContradictionRecords are appended symmetrically to both artefacts’ ledger entries. A ClaimExtractionManifest binds all extracted claims and contradiction records to the artefact. The operation executes after cross-domain stitching and before P94 monitor evaluation.