AIEP — Deterministic Iterative Evidence Refinement and Termination Certification System
Field of the Invention
[0001] The invention relates to deterministic iterative evaluation mechanisms within distributed computing systems.
[0002] More particularly, the invention concerns deterministic intra-cycle evidence refinement and termination certification within an Architected Instruction & Evidence Protocol (AIEP) system as defined in United Kingdom patent application GB2519711.2.
Background
[0003] The AIEP system defined in GB2519711.2 performs deterministic arbitration of canonical artefacts under registry-bound scoring semantics and fail-closed admissibility gating.
[0004] In certain deployments, arbitration may require iterative refinement of artefacts prior to final admissibility determination.
[0005] Conventional iterative systems frequently rely on heuristic stopping criteria, variable convergence thresholds, or discretionary human intervention.
[0006] Such approaches may produce non-reproducible iteration counts, non-deterministic convergence states, and inconsistent admissibility outcomes across distributed nodes.
[0007] There exists a need for a deterministic iterative refinement mechanism that:
(a) enforces strictly ordered iteration progression;
(b) binds each iteration to a verifiable state hash;
(c) uses registry-defined termination conditions; and
(d) guarantees identical termination state under identical inputs.
Summary of the Invention
[0008] The invention provides a Deterministic Iterative Evidence Refinement and Termination Certification System (DIERTCS) operating as an intra-arbitration control layer within AIEP.
[0009] The system comprises:
(a) a deterministic iteration controller;
(b) a canonical input-output binding module;
(c) a lineage hash chaining engine;
(d) a registry-bound termination evaluator;
(e) a refinement certification generator; and
(f) a fail-closed admissibility gate controller.
[0010] The system operates within a single arbitration cycle and does not modify cross-cycle topology invariants defined elsewhere in the AIEP architecture.
Detailed Description
Deterministic Iteration Indexing
[0011] Each refinement cycle is assigned a strictly monotonic iteration index i beginning at i = 0.
[0012] Iteration increments are deterministic and sequential.
[0013] Skipped indices, parallel discretionary branching, or non-monotonic iteration are prohibited within a single refinement sequence.
Canonical Input–Output Binding
[0014] For each iteration i, input artefacts are canonicalised under active schema and registry versions.
[0015] Deterministic transformation rules are applied.
[0016] A canonical output artefact Oᵢ is generated.
[0017] Output artefacts are not permitted to overwrite prior artefacts; instead, each iteration produces a new artefact appended to lineage.
Iteration Lineage Hash Chaining
[0018] Each iteration state is bound to a deterministic hash computed as:
Hᵢ = H(Hᵢ₋₁ || CanonicalInputᵢ || CanonicalOutputᵢ || RegistryVersion || SchemaVersion || i)
[0019] The hash Hᵢ uniquely identifies the full state of iteration i.
[0020] Identical inputs under identical registry and schema versions yield identical Hᵢ sequences across distributed nodes.
Registry-Bound Termination Conditions
[0021] Termination conditions are defined in a versioned registry referenced in GB2519711.2.
[0022] Termination conditions may include:
(a) satisfaction of admissibility completeness criteria;
(b) convergence within defined deterministic tolerance bounds;
(c) exhaustion of permitted iteration count;
(d) detection of non-progress state; or
(e) explicit registry-defined stopping condition.
[0023] Termination criteria are immutable for a given registry version.
[0024] Heuristic or discretionary stopping conditions are prohibited.
Deterministic Loop Execution
[0025] For each iteration i:
(a) canonical input is evaluated;
(b) deterministic transformation rules are applied;
(c) canonical output Oᵢ is generated;
(d) iteration hash Hᵢ is computed; and
(e) termination conditions are evaluated.
[0026] If termination conditions are not satisfied, iteration index increments and the process repeats.
Refinement Certification
[0027] Upon termination, the system generates one of:
(a) a Refinement Completion Certificate (RCC); or
(b) a Refinement Exhaustion Certificate (REC).
[0028] The RCC indicates termination due to satisfaction of admissibility criteria.
[0029] The REC indicates termination due to exhaustion or registry-defined stopping condition without admissibility satisfaction.
[0030] Each certificate comprises:
(a) final iteration index;
(b) final iteration hash Hₙ;
(c) registry version identifier;
(d) schema version identifier; and
(e) cryptographic commitment H(Certificate).
[0031] Certificates are canonical artefacts appended to the AIEP lineage.
Fail-Closed Enforcement
[0032] Failure of canonicalisation, iteration indexing integrity, hash chaining, registry resolution, or certificate generation causes deterministic transition to a non-admissible and non-executable state.
[0033] No execution enablement signal may be generated in absence of a valid RCC.
Distributed Deterministic Equivalence
[0034] Where canonical inputs, registry versions, schema versions, and transformation rules are identical, independent computing systems produce identical iteration sequences, identical hash lineage values, and identical termination certificates.
[0035] The system thereby guarantees deterministic intra-cycle refinement equivalence across distributed AIEP nodes.
Technical Effect
[0036] The invention converts iterative refinement from heuristic convergence into a deterministic, registry-bound state-transition process.
[0037] The invention guarantees reproducible iteration counts, reproducible termination states, and verifiable refinement lineage across distributed systems.
[0038] The technical contribution lies in enforcing deterministic loop control, hash-bound state chaining, and fail-closed termination certification within a structured arbitration protocol.
CLAIMS
1. A computer-implemented method operating within an Architected Instruction & Evidence Protocol (AIEP) system as defined in United Kingdom patent application GB2519711.2, comprising:
(a) assigning a deterministic iteration index to each refinement cycle;
(b) canonicalising input artefacts under versioned schema and registry rules;
(c) generating canonical output artefacts for each iteration;
(d) computing a chained iteration hash binding prior state, input, output, registry version, schema version, and iteration index;
(e) evaluating registry-defined termination conditions;
(f) generating a refinement completion certificate or refinement exhaustion certificate; and
(g) enforcing fail-closed transition to a non-executable state upon failure of iteration integrity or termination certification.
2. The method of claim 1, wherein termination conditions include deterministic convergence thresholds or iteration-count limits defined in a versioned registry.
3. The method of claim 1 or 2, wherein identical inputs under identical registry and schema versions yield identical iteration hash sequences across distributed nodes.
4. A distributed computing system configured to perform the method of any preceding claim.
5. A non-transitory computer-readable medium storing instructions which, when executed, perform the method of any preceding claim.
Abstract
A deterministic iterative evidence refinement and termination certification control system operating within an Architected Instruction & Evidence Protocol (AIEP) distributed arbitration framework is disclosed. The system executes refinement as a deterministic state machine having a strictly monotonic iteration index. For each iteration, canonicalised input artefacts are transformed under versioned registry and schema rules to generate canonical outputs. Each iteration state is bound to a chained hash incorporating prior state, input, output, registry version, schema version, and iteration index. Termination conditions are defined exclusively by immutable, versioned registry invariants. Upon termination, a refinement completion certificate or refinement exhaustion certificate is generated and cryptographically bound to the final iteration hash. Execution enablement within the AIEP system is permitted only upon verification of a valid refinement completion certificate. The system thereby guarantees reproducible iteration counts, identical termination states, and deterministic execution gating across distributed computing nodes operating under identical inputs and registry conditions.
Brief Description of the Drawing
FIG. 1 — Deterministic Processing Pipeline
┌───────────────┐ InputType ┌─────────────────────┐
│ Input Object │──────────────▶│ NormalisationProfile│
│ (any format) │ │ (version-bound) │
└───────────────┘ └──────────┬──────────┘
│ parsing rules
┌──────────▼──────────┐
│ Processing Engine │
│ • encoding rules │
│ • ordering rules │
│ • lossless check │
│ • fail-closed gate │
└──────────┬──────────┘
┌───────────────┴──────────────┐
┌────────▼────────┐ ┌─────────▼────────┐
│ CanonicalForm │ │ Rejection Record │
│ H(CF‖ProfileID)│ │ (violated rule) │
└────────┬────────┘ └──────────────────┘
│
┌────────▼────────┐
│ Canonical Ledger│
│ (append-only) │
└─────────────────┘