MetaForgeX AI · NSDM Institute · Independent research programme

Governed AI for decisions, actions and infrastructure.

MetaForgeX builds provider-neutral intelligence that determines whether an AI workload, decision or action is supported, authorised, compliant and worth its cost—before execution and throughout runtime.

From evidence-bound decision systems to sovereign workload routing and outcome economics, we make AI operationally defensible. Capability is not permission, and low cost is not value unless the result is correct, accepted and authorised.

Explore the research Discuss a deployment What is an NSDM?
NSDM evidence and governance states evaluated at a justification boundary before admissible action
The model may be capable of answering, recommending or acting. NSDM determines whether it is justified and permitted to do so.

The control layer AI infrastructure is missing

Cloud providers expose compute. Models produce outputs. Cost platforms report spend. MetaForgeX connects the unresolved questions: which evidence supports the decision, whether the actor has authority, where a workload may run, which route is justified, whether the outcome was accepted, and whether the full action can be audited.

NSDM

Evidence-bound decisions

Determine whether a claim is supportable, whether a decision remains justified and whether an action is permitted under evidence, uncertainty, authority, policy and consequence.

AI Efficiency OS

Governed outcome economics

Know what every AI task costs—and whether that expenditure produced a correct, authorised, accepted and deployable outcome.

Sovereign AI

Federated workload intelligence

Determine where a workload may legally run, which provider can execute it, which model and hardware route is justified, and how the allocation is audited.

Neurosymbolic AI

Capability becomes governed execution.

Neural AI recognises patterns in language, images, behaviour, sound, video and data. Symbolic AI represents rules, definitions, permissions and logical relationships.

NSDM turns that combination into a decision machine: it checks whether evidence is sufficient, whether the world model is reliable, whether authority and jurisdiction are clear, what the consequence of error would be, and whether the system should answer, act, abstain, reroute or escalate.

Neural perceptionInterpret uncertain language, images, telemetry, behaviour or other signals.
Symbolic representationMap facts into explicit concepts, rules and relationships.
World-state checkingTest whether the system's representation of the situation is coherent and valid enough for the task.
Constraint checkingTest the proposal against evidence, policy, logic, permissions and known structure.
NSDM boundary evaluationLabel the evidence-state, governance-state, consequence level and permitted action.
Controlled responseAnswer, generate, recommend, simulate, ask, verify, abstain, refuse or escalate.
Computational experimentation and visual evidence

Making machine decisions visible.

NSDM uses Python, MATLAB, Wolfram, computer vision, symbolic computation, simulation and animated diagnostics to reveal what a model learned, where its decision boundary lies, how uncertainty changes, and why a proposed action passes or fails its evidence and governance checks.

NSDM evidence, governance and consequence states routing through a justification boundary to admissible or non-admissible actions
Visualisations are treated as research artefacts, not decoration. Each graph, animation, scene model or symbolic derivation must connect to an experiment, a dataset, a model version and a claim boundary—showing what is measured, what is inferred and what the result does not prove.
Working mathematical frame

Capability is not the same as admissibility.

Justified(A) is true only when E + R + C + K are sufficient for the action.
The greater the consequence of error, the stronger the evidence, control, and escalation requirements.
A — actionE — evidenceR — rules and authorityC — context and world stateK — consequence

Research is becoming executable governance infrastructure

NSDM now connects formal research to an executable governance kernel. The first accepted Workbench milestone separates authenticated identity from explicit Case-initiation authority and creates a tenant-scoped Case, AssessmentVersion v1 and CASE_CREATED AuditEvent as one governed transaction. Governed evidence, decision synthesis and Action Assurance remain subsequent milestones.

NSDM Institute research and product topology
Research becomes infrastructure: datasets, benchmarks, evidence objects, deployment gates, and vertical solutions.

Research papers

The paper programme is separated into working papers and planned manuscripts. No draft is presented as peer-reviewed or complete.

Working paper

Decision Boundaries and Benchmark

Core theory, labels, benchmark design, and evidence/action architecture.

View paper programme →
Research track

Neurosymbolic Decision Systems

Neural perception, symbolic constraints, evidence states, governance states, and controlled action.

View paper programme →
Research track

Sovereign Workload Boundaries

Jurisdiction, data residency, provider capability, model placement, energy, cost, authority and human override.

View paper programme →

How people use NSDM

NSDM turns theory and benchmark data into concrete product, research, governance, and deployment work.

Researchers

Build better benchmarks

Classify failure cases, measure calibration, compare boundary behaviour, and publish reproducible findings.

AI builders

Encode safer behaviour

Add evidence checks, abstention, refusal, escalation, and human review into product flows.

Enterprises

Deploy with proof

Use evidence ledgers, governance profiles, and deployment gates before scaling AI.

Boards and regulators

Challenge claims

Inspect authority, accountability, contestability, cost, and consequence.

See role-based workflows Explore verticals

How I make this

The way I would want any lab to disclose its methods: openly.

I use AI in the making of all MetaForgeX AI products and services, the way I use a spreadsheet or a search engine. The arguments, angles, opinions, and scars are mine—earned across decades spent building and interrogating technology, AI, enterprise architecture, research, neuroscience, neuromarketing, sponsorship, and marketing.

AI helps me research, pressure-test, and draft; I direct, cut, rewrite, and sign off on every word.

Every fact is verified against primary sources before it ships. Nothing is fabricated. Where a claim is provisional, I label it.

— The NeuroAlchemist, MetaForgeX AI

Programme status - 12 August 2026: The first executable NSDM Workbench milestone is accepted. M1 - Governed Case Kernel separates authenticated identity from explicit Case-initiation authority; successful initiation creates a tenant-scoped Case, AssessmentVersion v1 and CASE_CREATED AuditEvent atomically, while missing or suspended authority is refused with zero partial records. This is an engineering milestone, not evidence that later governed-evidence, decision-synthesis or Action Assurance capabilities are complete. The public research record continues to preserve positive, partial and null findings, including the EXP-001 adjudication result and the EXP-008 prospective null result.

Build AI that can justify where, why and at what cost it acts.

MetaForgeX AI develops the governance and economics control layer between AI intent and permitted execution: evidence-bound decisions, authorised actions, sovereign workload routing and cost-to-successful-outcome intelligence.

Organisation

MetaForgeX AI (Pty) Ltd

Johannesburg, South Africa