UK Patent Application
Applicant: Neil Grassby
AIEP --- Recursive Goal Tree Deepening Architecture with Plausibility-Bounded Termination
Abstract
A computer-implemented method and system for recursive generation of sub-goals from unresolved GoalVector search operations within a governed distributed reasoning substrate. When a directed search initiated by a GoalVector encounters irresolvable incompatibilities or evidence gaps, the system automatically generates sub-goal nodes in a deterministic goal tree structure. Each sub-goal inherits lineage from its parent GoalVector and carries a SubGoalHash encoding its derivation. Recursive deepening is bounded by a plausibility-matrix-enforced termination condition preventing unbounded goal tree expansion. The invention provides deterministic, lineage-intact goal tree growth as an architectural consequence of evidential search.
Description
Field of the Invention
[0001] The present invention relates to governed distributed reasoning substrates and more particularly to recursive generation of directed sub-goals from unresolved goal search operations within an Architected Instruction & Evidence Protocol (AIEP) system.
Background to the Invention
[0002] Governed reasoning substrates maintain GoalVectors representing directed search objectives generated from sustained incompatibility between co-active reasoning branches as described in relation to prior AIEP filings.
[0003] A directed search toward a GoalVector may encounter conditions in which the incompatibility cannot be resolved with currently available evidence, or in which the search reveals further specific incompatibilities whose resolution is a precondition for resolving the original goal.
[0004] Existing systems do not provide a deterministic mechanism for automatically generating sub-goals from these search conditions whilst preserving full lineage integrity and enforcing bounded termination.
[0005] There exists a need for a recursive goal tree deepening architecture that generates sub-goals deterministically from unresolved search conditions, preserves hash-bound lineage from sub-goal to root GoalVector, and enforces plausibility-bounded termination.
Summary of the Invention
[0006] According to a first aspect of the invention, there is provided a computer-implemented method for recursive goal tree deepening in a governed reasoning substrate, the method comprising:
(a) conducting a directed search toward an active GoalVector;
(b) detecting a deepening condition comprising one or more of: insufficient evidence to resolve the GoalVector incompatibility; identification of a specific sub-incompatibility whose resolution is prerequisite to GoalVector resolution; or a search gap encoding a defined evidence deficit;
(c) automatically generating a sub-goal node in the goal tree, the sub-goal node encoding the specific deepening condition;
(d) computing a SubGoalHash as a deterministic cryptographic function over: the parent GoalVector hash, the deepening condition encoding, and a generation timestamp;
(e) inserting the sub-goal node into the substrate with directed lineage edges to the parent GoalVector; and
(f) evaluating sub-goal plausibility against the active plausibility matrix and suppressing generation in a fail-closed manner upon plausibility violation.
[0007] According to a second aspect, there is provided a system comprising: a search engine configured to conduct directed searches toward active GoalVectors; a deepening condition detector; a sub-goal generation controller; a SubGoalHash computation module; and a plausibility termination enforcer.
[0008] According to a third aspect, there is provided a non-transitory computer-readable medium storing instructions which, when executed, cause a processor to perform the method of the first aspect.
Brief Description of the Drawings
Figure 1 illustrates a goal tree showing root GoalVector with sub-goal nodes.
Figure 2 illustrates deepening condition detection and sub-goal generation.
Figure 3 illustrates SubGoalHash computation and lineage insertion.
Figure 4 illustrates plausibility-bounded termination enforcement.
Detailed Description of Preferred Embodiments
1. GoalVector Directed Search Context
[0009] The invention operates within a governed reasoning substrate maintaining active GoalVectors generated from sustained incompatibility between co-active reasoning branches.
[0010] A directed search toward a GoalVector comprises evidence retrieval, branch evaluation, and resolution assessment operations conducted under constitutional governance constraints.
[0011] The directed search terminates under one of three conditions: resolution, deepening, or transformation.
[0012] The present invention governs the deepening condition.
2. Deepening Condition Detection
[0013] A deepening condition is detected when the directed search determines that:
(a) currently available evidence is insufficient to resolve the GoalVector incompatibility and a specific evidence gap can be identified;
(b) a sub-incompatibility exists whose resolution is a necessary precondition for GoalVector resolution; or
(c) the search scope reveals a distinct irresolvable tension at a more specific level than the parent GoalVector.
[0014] Deepening condition detection is deterministic and schema-governed.
[0015] The deepening condition is encoded as a structured descriptor comprising: the type of condition, the specific incompatibility or gap identified, and references to the evidence artefacts establishing the condition.
3. Sub-Goal Node Generation
[0016] Upon detection of a deepening condition, the sub-goal generation controller automatically creates a sub-goal node without external instruction.
[0017] The sub-goal node comprises:
(a) a sub-goal identifier;
(b) a deepening condition encoding;
(c) a SubGoalHash;
(d) a parent GoalVector reference;
(e) a generation timestamp;
(f) a plausibility attestation.
[0018] The sub-goal node is inserted into the substrate DAG with directed edges to the parent GoalVector node.
[0019] Multiple sub-goals may be generated from a single GoalVector search where multiple distinct deepening conditions are detected.
4. SubGoalHash Computation
[0020] A SubGoalHash is computed as a cryptographic hash over:
(a) the GoalVectorHash of the parent GoalVector;
(b) a canonical serialisation of the deepening condition encoding; and
(c) a generation timestamp.
[0021] SubGoalHash is incorporated into sub-goal node identity.
[0022] Identical deepening conditions detected from identical parent GoalVectors produce identical SubGoalHash values across distributed nodes.
[0023] The full lineage path from any sub-goal to its root GoalVector is computable via hash traversal.
5. Plausibility-Bounded Termination
[0024] Each generated sub-goal is evaluated against the active plausibility matrix prior to insertion.
[0025] Sub-goal generation is suppressed fail-closed when the sub-goal encoding falls outside the plausibility space defined by the active plausibility matrix.
[0026] Maximum goal tree depth is enforced as a schema-defined parameter.
[0027] The plausibility termination condition prevents unbounded recursive goal tree expansion whilst preserving all legitimate deepening paths.
[0028] A plausibility termination record is appended to lineage upon suppression, recording the sub-goal hash and the plausibility constraint violated.
Technical Effect
[0029] The invention provides deterministic recursive generation of directed sub-goals from evidential search operations.
[0030] It preserves full hash-bound lineage from every sub-goal to its root GoalVector.
[0031] It prevents unbounded goal tree expansion via plausibility-enforced termination.
[0032] It enables machine-speed recursive problem decomposition as an architectural consequence of evidential search.
CLAIMS
1. A computer-implemented method for recursive goal tree deepening in a governed reasoning substrate, the method comprising: conducting a directed search toward an active GoalVector; detecting a deepening condition comprising identification of an irresolvable sub-incompatibility or evidence gap within the directed search; automatically generating a sub-goal node encoding the deepening condition without external instruction; computing a SubGoalHash as a cryptographic function over the parent GoalVector hash, the deepening condition encoding, and a generation timestamp; inserting the sub-goal node into the substrate with directed lineage edges to the parent GoalVector; and suppressing sub-goal generation fail-closed upon determination that the sub-goal encoding falls outside the active plausibility matrix.
2. The method of claim 1 wherein the deepening condition comprises insufficient evidence to resolve the GoalVector incompatibility with a specific evidence gap identified.
3. The method of claim 1 wherein the deepening condition comprises identification of a sub-incompatibility whose resolution is a necessary precondition for GoalVector resolution.
4. The method of claim 1 wherein identical deepening conditions detected from identical parent GoalVectors produce identical SubGoalHash values across distributed nodes.
5. The method of claim 1 wherein the full lineage path from a sub-goal to its root GoalVector is computable via hash traversal of SubGoalHash values.
6. The method of claim 1 wherein maximum goal tree depth is enforced as a schema-defined parameter.
7. A system for recursive goal tree deepening in a governed reasoning substrate, comprising: a directed search engine; a deepening condition detector configured to identify irresolvable sub-incompatibilities or evidence gaps; a sub-goal generation controller; a SubGoalHash computation module; and a plausibility termination enforcer configured to suppress sub-goal generation upon plausibility violation.
8. A non-transitory computer-readable medium storing instructions which, when executed, perform the method of claim 1.
Abstract
A computer-implemented method and system for recursive generation of sub-goals from unresolved GoalVector directed searches within an AIEP governed reasoning substrate. When a directed search toward an active GoalVector encounters an irresolvable sub-incompatibility or evidence gap, a sub-goal node is automatically generated encoding the deepening condition. A SubGoalHash is computed over the parent GoalVector hash, deepening condition encoding, and timestamp, preserving full hash-bound lineage to the root GoalVector. Sub-goal generation is evaluated against the active plausibility matrix and suppressed fail-closed upon plausibility violation, preventing unbounded recursive expansion. The invention produces deterministic goal tree growth as an architectural consequence of evidential search.
Brief Description of the Drawing
FIG. 1 — GoalVector Stabilisation Architecture
┌──────────────────────────┐
│ Constitutional Goal │
│ G_root (immutable) │
└───────────┬──────────────┘
│ decompose
┌───────────┼───────────┐
┌────▼────┐ ┌──▼──────┐
│ SubGoal │ │ SubGoal │
│ G1 │ │ G2 │
└────┬────┘ └──┬──────┘
│ │
┌────▼────┐ ┌─────▼────┐
│ Action │ │ Drift? │
│ Plan A1 │ │ YES─────┼──▶ re-anchor
└─────────┘ │ NO ─────┼──▶ continue
└──────────┘
Stability = 1 − GoalDriftMagnitude / MaxPermittedDrift