P232 — AIEP — Knowledge Utility Scoring Engine

Applicant: Neil Grassby Classification: Patent Application — Confidential Priority: Claims priority from GB2519711.2 filed 20 November 2025 Architecture Layer: AIEP Phase 2 Support Layer

Framework Context

[0001] This specification operates within an AIEP environment as defined in GB2519711.2 and GB2519798.9. The present specification defines a mechanism for computing utility scores for knowledge items in the AIEP knowledge stores, enabling the Knowledge Distillation Engine (P217) and Long-Term Reasoning Memory Engine (P208) to prioritise high-value knowledge for retention and distillation.

Field of the Invention

[0002] The present invention relates to knowledge value assessment systems for evidence-bound artificial intelligence knowledge management.

[0003] More particularly, the invention relates to a multi-criterion knowledge utility scoring system that evaluates evidence artefacts, CWSG nodes, abstracted inference rules, and distilled claims on dimensions including reasoning frequency, goal relevance, evidence uniqueness, and temporal currency.

Background

[0004] Evidence-bound AI systems accumulate large volumes of evidence artefacts, world state nodes, and reasoning records over time. Not all knowledge items have equal value for future reasoning tasks. Without a utility scoring mechanism, knowledge stores grow without bound, retrieval latency increases, and the most valuable items may be buried beneath low-utility records.

Summary of the Invention

[0005] The invention provides a Knowledge Utility Scoring Engine (KUSE) that assigns a utility score to each trackable knowledge item by evaluating: reasoning frequency (how often the item has been retrieved in reasoning sessions); goal contribution (how frequently the item contributed to goal completion); evidence uniqueness (fraction of claims supported by this item that are not supported by other items); temporal currency (recency and freshness of the underlying evidence); and cross-reference weight (citation count from other knowledge items).

[0006] Utility scores are computed on each scoring cycle and stored as metadata on the knowledge item. Items with scores below the retention threshold are flagged for archival; items with scores above the distillation threshold are flagged for distillation by P217.

ASCII Architecture

Knowledge Stores:
  Evidence Artefacts
  CWSG Nodes
  Abstracted Rules (P231)
  LTM Records (P208)
           |
           v
+------------------------------------------+
| Knowledge Utility Scoring Engine (KUSE)  |
|                                          |
|  Reasoning Frequency Counter            |
|  Goal Contribution Calculator           |
|  Uniqueness Analyser                    |
|  Temporal Currency Assessor             |
|  Cross-Reference Counter                |
+-------------------+----------------------+
                    |
                    v
   Utility Score per Item
   → Archive Flag (score < lower_threshold)
   → Distillation Flag (score > upper_threshold)

Detailed Description

[0007] Reasoning Frequency. The KUSE maintains a usage counter for each knowledge item, incrementing it each time the item is retrieved in a reasoning session. Frequency is normalised by total sessions in the scoring window to produce a frequency score.

[0008] Goal Contribution. The KUSE tracks which knowledge items were cited in reasoning sessions that resulted in goal completions. Items consistently cited in successful goal-advancing sessions receive elevated goal contribution scores.

[0009] Evidence Uniqueness. For each evidence artefact, the KUSE assesses the fraction of the claims it supports that are not corroborated by other artefacts. Unique primary evidence has higher utility than heavily duplicated secondary evidence.

[0010] Temporal Currency. Items derived from recent evidence artefacts receive higher temporal currency scores than items derived from older, potentially outdated evidence. Staleness flags from the ESR and CWSG staleness evaluators reduce temporal currency scores.

[0011] Composite Utility Score. The composite utility score is computed as: U = w1 * frequency_score + w2 * goal_contribution + w3 * uniqueness_score + w4 * temporal_currency + w5 * cross_reference_weight. Default weights: w1=0.25, w2=0.30, w3=0.20, w4=0.15, w5=0.10.

[0012] Archival and Distillation Flags. Items scoring below lower_threshold (default 0.15) are flagged for archival — moved to cold storage but not deleted. Items scoring above upper_threshold (default 0.70) are flagged for distillation by P217.

Technical Effect

[0013] The invention provides evidence-grounded knowledge prioritisation that enables efficient archival and distillation decisions for large AIEP knowledge bases. By computing utility scores across multiple independent dimensions — reasoning frequency, goal contribution, evidence uniqueness, temporal currency, and cross-reference density — the engine reflects the true epistemic value of knowledge items rather than relying on single-dimension proxies. By automatically flagging high-utility items for distillation and low-utility items for archival, the engine maintains a high-value active knowledge base without manual curation.

Claims

1. A computer-implemented method for knowledge utility scoring, the method comprising: (a) computing a reasoning frequency score for each knowledge item based on citation count in reasoning sessions within a policy-defined rolling window; (b) computing a goal contribution score based on citation frequency in sessions resulting in goal completion events; (c) computing an evidence uniqueness score as the proportion of claims supported by the item that are not corroborated by other evidence artefacts; (d) computing a temporal currency score incorporating staleness flags from the entity state registry and evidence age relative to the current CWSG timestamp; (e) computing a composite utility score as a weighted sum of component scores; and (f) flagging items below a lower utility threshold for archival and items above an upper utility threshold as distillation candidates.

2. The method of claim 1, wherein component score weights are configurable via the active governance policy.

3. The method of claim 1, wherein archival-flagged items are moved to cold storage without deletion, preserving access for historical reasoning queries.

4. The method of claim 1, wherein distillation-flagged items are forwarded to the Knowledge Distillation and Compression Engine as high-utility distillation candidates.

5. The method of claim 1, wherein utility score updates are recalculated on each evidence admission cycle rather than on a fixed schedule, ensuring scores reflect current reasoning activity.

6. A Knowledge Utility Scoring Engine comprising: one or more processors; memory storing a citation index, goal completion citation log, and utility score store; wherein the processors are configured to execute the method of claim 1.

7. A non-transitory computer-readable medium storing instructions that, when executed by a processor, implement the method of claim 1.

Abstract

A knowledge utility scoring engine for evidence-bound artificial intelligence computes composite utility scores for knowledge items across five independent dimensions: reasoning citation frequency, goal contribution, evidence uniqueness, temporal currency, and cross-reference density. Items falling below a lower utility threshold are flagged for archival; items exceeding an upper threshold are flagged for distillation by the Knowledge Distillation Engine. Score updates are recalculated on each evidence admission cycle to reflect current reasoning activity.