Control before execution

Technology

SentientROUTER is the control layer. It stands before the actions of AI and makes authority visible to the people who operate and review the system.

Architecture

Policy is separate from model output

A model produces a result. Policy decides what happens with it. SentientROUTER keeps these two things apart.

Policy

Rules and versions

Each classification names the policy version and its scope.

Authority

Human decision points

A high-impact action waits until the authorised person records a decision.

State

Conservative escalation

Divergence, missing context or a governance concern narrows automation. Nothing widens it silently.

Evidence

Events for review

The classification and its context pass to an evidence layer for later checks.

Mechanisms

Four mechanisms of control

  • Constitution first

    The rules are compiled into the program as a constitution with a version. SentientROUTER checks each request against it before any autonomous action and before any external source is queried.

  • Consensus of sources

    A consensus decision uses at least three independent sources. The system records each vote with its hash. No single source decides alone.

  • Drift detection

    SentientROUTER watches how often requests go to S, D and H, how far the sources diverge and how the control state varies. Drift narrows authority.

  • Work without a link

    Control continues through a loss of connectivity. The records synchronise when the link returns.

AI models

One governed route to different AI models

SentientROUTER gives one common route to different AI models. The same request rules apply to each model and each tool.

The gateway contains
Provider adapters, request processing and checks of the execution contract.
Each use case states
Its data access, its permitted providers and its behaviour on failure.

Control and proof

Control and proof have different roles

SentientROUTER decides whether an AI action may proceed. ProvableCORE makes a record of each decision.

The control layer

SentientROUTER

It routes a request under the rules of the organisation and human authority.

The proof layer

ProvableCORE

It binds each decision to its context, its policy, its approvals and its integrity evidence. The record is signed. A third party can verify it.

provablecore.com

In the SoulGCI core

One closed control loop

SoulGCI is the governed intelligence core of the group. In the core, SentientROUTER is the control layer and ProvableCORE is the proof layer.

Human authoritySets the limits. Approves or stops.

Control layer

SentientROUTERIs this action permitted?

Governed execution

Permitted actionThe action runs inside its limits.

Proof layer

ProvableCOREWhat happened, and under which rules?
Verified results set the limits of future autonomy.

The seven steps of the core

  1. Take in data
  2. Link the data
  3. Build understanding
  4. Form a decision
  5. Check authority
  6. Act and record
  7. Check and learn

Step 5 is the work of SentientROUTER: act automatically, stop for review, or wait for approval.

The SoulGCI core

Institutional context

Designed for institutional model governance

Human approval, traceability and effective review matter to institutions in the Americas. SentientROUTER is designed with these three needs in mind. Each deployment has its own legal and risk assessment.

  • Defined responsibility

    The institution determines which roles approve a consequential action, and under which policy.

  • Reviewable history

    A reviewer sees the policy, the class and the human decision for each event.

Questions

What organisations ask

Does SentientROUTER replace the approval process of an organisation?

No. The approval rules stay with the organisation. SentientROUTER holds the permitted route for each request and hands the request to the authorised people.

Can the confidence of a model increase its autonomy?

No. Confidence does not widen delegated authority. Autonomy changes only through an explicit policy and an approval process.

How does an organisation start to use SentientROUTER?

Each use case starts with an assessment of the technical controls, the security, the legal obligations and the integrations. Then the organisation defines the policy, the thresholds, the authorised roles and the evidence requirements.

How does ProvableCORE fit in?

ProvableCORE is the proof layer. SentientROUTER decides whether an action may proceed. ProvableCORE makes a signed record of the decision, and a third party can verify the record.