P255 — AIEP — Proof-of-Reconstitution Verification Engine

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

Framework Context

[0001] This specification operates within an AIEP environment as defined in GB2519711.2 and GB2519798.9. The present specification defines the verification engine that processes reconstitution proof records (P237) to confirm that distilled knowledge claims and compressed lineage records remain faithfully derived from their original evidence artefacts.

Field of the Invention

[0002] The present invention relates to proof-of-reconstitution verification for distilled knowledge and compressed reasoning lineage in evidence-bound AI.

Background

[0003] Over time, the evidence corpus is distilled and compressed to manage storage and retrieval efficiency. Without active verification, distilled claims may silently diverge from their source evidence due to artefact storage corruption, parameter drift, or administrative modification. The Proof-of-Reconstitution Verification Engine provides continuous active assurance that distilled knowledge is traceable to source evidence.

Summary of the Invention

[0004] The invention provides a Proof-of-Reconstitution Verification Engine (PRVE) that: maintains a verification schedule covering all distilled claims and compressed lineage records with active reconstitution proofs (P237); periodically re-runs the reconstitution verification algorithm (re-applying distillation to source artefacts and confirming output hash match); flags verification failures as integrity alerts; and updates verification status metadata on each knowledge item.

[0005] The PRVE supports two verification modes: spot check (random sample verification on each cycle) and full audit (complete verification of all items with active proofs, triggered by governance events or on operator demand).

ASCII Architecture

Reconstitution Proof Records (P237)
         |
         v
+------------------------------------------+
| Proof-of-Reconstitution Verification     |
|   Engine (PRVE)                          |
|                                          |
|  Verification schedule management       |
|  Source artefact retrieval              |
|  Distillation function replay           |
|  Output hash comparison                 |
|  Verification status update             |
+-------------------+----------------------+
                    |
            Verification result:
            PASS → Status updated
            FAIL → Integrity alert → Governance escalation

Detailed Description

[0005] Verification Schedule. The PRVE maintains a scheduling table assigning a next-verification timestamp to each knowledge item with an active proof. Default verification interval is configurable; high-utility items (P232) are verified more frequently than low-utility items.

[0006] Source Artefact Retrieval. On each verification, the PRVE retrieves the source artefacts from the Merkle tree recorded in the proof record, verifying artefact integrity against stored hashes before distillation replay.

[0007] Distillation Replay. The same distillation function version and parameters recorded in the proof record are applied to the source artefacts. The output hash is computed.

[0008] Comparison and Status. If the replay output hash matches the certified claim hash in the proof record, the verification passes and the last_verified timestamp on the knowledge item is updated. If hashes diverge, an integrity alert is raised, the knowledge item is flagged UNVERIFIED, and governance escalation is triggered.

Technical Effect

[0009] The invention provides ongoing, schedule-driven verification that distilled knowledge claims remain evidentially traceable to their source artefacts, detecting integrity failures caused by source artefact retraction, corruption, or evidence base modification after distillation. By replaying the original distillation function and parameters, the engine tests reconstitutability in situ rather than merely checking hash records, detecting cases where source artefacts have been replaced by differently-hashed artefacts with the same identifier. Utility-proportional scheduling concentrates verification resources on the highest-value knowledge items, providing a risk-proportionate assurance programme.

Claims

1. A method of proof-of-reconstitution verification for distilled knowledge items in an evidence-bound artificial intelligence system, comprising the steps of: (a) maintaining a verification schedule assigning a next-verification timestamp to each knowledge item with an active proof record, scheduling high-utility items more frequently based on utility scores produced by the Knowledge Utility Scoring Engine; (b) at each scheduled verification, retrieving the source artefacts referenced in the Merkle tree of the proof record and verifying each artefact’s content hash against the hash stored in the proof record before proceeding; (c) applying the distillation function version and parameters recorded in the proof record to the retrieved source artefacts and computing the output hash; (d) on hash match between the replay output hash and the certified claim hash in the proof record, updating the last_verified timestamp on the knowledge item; (e) on hash divergence, flagging the knowledge item UNVERIFIED, raising an integrity alert, and triggering governance escalation via the Governance Safety Constraint Engine.

2. The method of claim 1, wherein source artefact integrity verification at step (b) uses the Merkle membership proof stored in the proof record to confirm the artefacts remain admitted members of the evidence ledger.

3. The method of claim 1, wherein spot check mode initiates a single on-demand verification of a specified knowledge item outside the regular schedule.

4. The method of claim 1, wherein full audit mode processes all knowledge items with active proofs in scheduled sequence, producing a verifiable audit completion record upon conclusion.

5. The method of claim 1, wherein a governance escalation triggered at step (e) includes the identity of the knowledge item, the source artefacts that failed hash verification, and the session in which the distillation was originally certified.

6. A proof-of-reconstitution verification engine for an evidence-bound artificial intelligence system, comprising: a verification schedule manager assigning next-verification timestamps proportional to knowledge utility scores; a source artefact retrieval and integrity checker; a distillation replay executor; a hash comparison module; and an alert and governance escalation dispatcher.

7. A computer-readable medium carrying instructions for implementing the method of any preceding method claim.

Abstract

A proof-of-reconstitution verification engine for evidence-bound artificial intelligence maintains a utility-proportional verification schedule for distilled knowledge items, periodically retrieving the source artefacts referenced in each item’s proof record and replaying the original distillation function and parameters. The replay output hash is compared to the certified claim hash in the proof record; on match, the last_verified timestamp is updated; on divergence, the knowledge item is flagged UNVERIFIED and governance escalation is triggered. Spot check and full audit modes supplement the scheduled programme.