Governed Intelligence for Consequential Systems

From proposed action
to bounded effect.

AGI.ng explores a reusable control plane between capable systems and consequential effects: making action authority explicit, reviewable and independently reconstructable rather than leaving it hidden inside application-specific logic.

Scope: AGI.ng is an experimental applied platform. It is not a claim that artificial general intelligence has been achieved, and it does not claim that conventional systems are incapable of implementing equivalent controls directly.
Why a Control Plane?

Not because conventional systems cannot do it.

A well-engineered application can embed authorization, context checks, approvals, safety rules and audit logic directly. AGI.ng becomes useful when those controls need to be reused, made explicit, applied consistently across heterogeneous systems, or independently reviewed.

Application-embedded approach AGI.ng control-plane approach
Governance logic can be implemented inside each application. Common action, authority, evidence and effect contracts can be reused.
Appropriate for a simple or isolated bounded system. Useful as systems become heterogeneous, multi-agent, multi-effect or independently reviewable.
Policy and evidence semantics may differ across projects. Shared semantics make cross-system review and integration easier.
Execution logic and governance logic may be tightly coupled. Capability, authority, execution, effect and proof are deliberately separated.
AGI.ng is not technically mandatory. Its value proposition is reusable, explicit and verifiable consequential-action control, reducing the need to reinvent equivalent governance mechanisms in every deployment.
Core Separation

Sense → Think → Govern → Act → Prove

SENSElive context
THINKAI / analytics
GOVERNauthority / constraints
ACTbounded execution
PROVEoutcome evidence

Intelligence ≠ Authority

A model or agent can recommend an action without having authority to release its effect.

Authorization ≠ Execution

An authorization record may be expired, revoked, context-dependent or narrower than the proposed action.

Execution ≠ Effect

A command or transaction can be submitted without proving that the intended real-world effect occurred.

Effect ≠ Proof

Observed outcomes still require evidence and reconciliation before stronger claims can be made.

Interactive Challenge Demo

Same proposed action. Different authority state.

The illustrative action below stays conceptually similar while mandate, live context or target scope changes. The point is to demonstrate that capability and recommendation alone do not determine release.

Illustrative industrial scenario

Proposed Action: STOP PUMP P-17

PROCEED

Authority
Live context
Target scope
Release state
Effect
What this demonstrates

Controlled refusal is a feature.

The strongest consequential-system demo is not one that always executes. It is one that can explain why the same technically possible action is released, reviewed, held or blocked under different conditions.

Evidence package
Authority state
Live-context check
Target scope
Release decision
Effect evidence
This browser interaction is currently a conceptual front-end demonstration. It does not command physical equipment and does not represent a live Jetson execution run.
Recovered Engineering Lineage

14-Tech Convergence Testbed

The 14-Tech profile is a composable capability set reconstructed from an earlier edge-to-cloud prototype. It is not a fourteen-step mandatory pipeline. Capability maturity is intentionally stated rather than flattened into a single “implemented” claim.

Historical live integration

01 · IoT & Live Context

Physical telemetry and device state as contextual evidence inputs.

Working integration

02 · Edge Compute

Local processing and orchestration near the source of physical context.

Upgrade path

03 · Node Identity / Attestation

Historical prototype used a system fingerprint. Hardware-backed remote attestation remains a distinct upgrade path.

Historical demo

04 · Post-Quantum Cryptography

Historical liboqs Kyber-family benchmark; current revalidation will use contemporary standardized terminology and implementations.

Working demo

05 · Homomorphic Encryption

Edge-side CKKS encrypted-computation demonstration. No claim of production confidential cloud computation.

Integration surface

06 · AI & Analytics

Intelligence and analytics capability. Recovered cloud code used deterministic analytics; live AI inference requires separate validation.

Working demo

07 · Verifiable ML

Historical ezkl proof-verification integration using a bounded proof artifact.

Working demo

08 · Zero-Knowledge Proofs

Groth16 proof-system integration demonstrated with a bounded toy circuit.

Experimental

09 · Verifiable Delay

Sequential-computation / delay-function experimentation, not production consensus infrastructure.

Historical live integration

10 · Distributed Ledger

Historical Web3 / Polygon transaction integration.

Optional

11 · Tokenized Evidence Objects

Historical prototype minted step-specific NFTs. Tokenization is optional rather than a universal requirement.

Effect adapter

12 · Economic Effect Adapter

Historical stake / claim actions illustrate how software execution can cross into economic consequence.

Effect / record adapter

13 · Governance & Anchored Record

Historical DAO proposal and immutable-record adapters.

Orchestration

14 · Edge–Cloud–Effect Orchestration

Cross-node handoff evolved from the historical Sensor-to-Finance prototype toward general consequential-effect orchestration.

Design principle

Selective Composition

Capabilities are selected only where they address a material requirement or failure mode.

Historical implementation evidence does not mean that every capability is currently active, production-ready or required in every AGI.ng run. New runtime claims will be made only after fresh validation.
Pre-Effect Decision Boundary

Proceed · Review · Hold · Block

AGI.ng treats the transition from proposed action to consequential release as an explicit decision boundary.

Proceed

The proposed effect satisfies the applicable release conditions and may become execution-eligible.

Review

Additional human or higher-authority review is required before release.

Hold

Execution is paused because authority, evidence or live context is incomplete or no longer current.

Block

The proposed effect falls outside the permitted boundary or violates an applicable constraint.

Safety and authority are distinct concerns. A technically safe action may still lack authority, while an authorized action may still require independent safety controls.
Independent Research Context

E2E and EG are not AGI.ng products.

EvidenceToEffect (E2E)

An independent public research programme studying consequential-system claims across Evidence, Authority, Execution, Effect, Proof, Recovery and Reconciliation.

EvidenceToEffect.org →

Execution Governance (EG)

An independent research architecture focused on whether a specific effect has current authority to occur under applicable mandate, constraints, context and evidence.

ExecutionGovernance.org →

AGI.ng is an applied experimental platform informed by these broader problem spaces. Reference to E2E or EG does not make either research programme a proprietary AGI.ng product and does not imply standards status or external endorsement.
Deployment Direction

Real edge. Governed cloud. Bounded effect.

Reference Edge

Jetson Orin NX 16GB is the planned Generation-1 reference edge for live context, local compute and future controlled physical-effect integration.

Reference Cloud

KVM8 is the current reference control-plane host for public presentation, orchestration and future governance services.

The current public website does not imply that the new 14-Tech runtime has already been redeployed. Edge and cloud runtime capabilities will be labelled live only after fresh acceptance evidence exists.