Document Type: Architecture Paper / Normative Reference Document

Context: Semantic Graph Layer · Knowledge Architecture

Status: Public Standard

Validity: Aivis-OS Core Architecture

Reference: Builds on the Entity Inventory specification and feeds the Transport-Safe Content Layer and all Machine Interface Projections.

Deterministic Meaning under Probabilistic Retrieval

1. Initial situation

Modern AI systems do not evaluate information deterministically, but probabilistically. “Truth” does not arise from individual facts, but from context density, relation types, and consistency.

An isolated entity has no authority for machines. Only the embedding in a coherent network of relationships, roles, and scopes makes information trustworthy.

The Semantic Graph Layer is the architectural answer to this reality.

2. Architectural principle

Graph Density as Authority

In the Aivis-OS architecture, authority does not arise from the mere existence of facts, but from the density, typing, and consistency of their links.

An isolated fact has little machine resilience. A graph-supported fact is stabilized by adjacent relations, responsibilities, and contexts.

The Semantic Graph Layer transforms the Entity Inventory from a collection of canonical objects into a traversable network of meanings.

3. From Inventory to Graph

The Entity Inventory answers the question:

What exists?

The Semantic Graph Layer answers another, more critical question:

What is valid – in what relationship, in what context, and with what reliability?

While the inventory manages entities, the graph manages statements about entities.

These statements are modeled as explicit relations and are therefore verifiable, versionable, and resolvable.

4. The basic problem: Ambiguity as punishment

AI systems punish ambiguity. If a Machine Interface Layer simultaneously exposes contradictory statements (e.g., two CEOs, two prices, two founding dates), there is no “weighing”, but a loss of trust at the domain level.

The model cannot decide which statement is valid and classifies the source as unreliable or outdated.

This leads to a central Aivis-OS principle: Determinism at the interface is not optional, but a prerequisite for trust.

5. The reality of organizational systems

However, organizations are not free of contradictions:

different data statuses
competing sources
temporal transitions
inconsistent responsibilities

A system that allows only a single truth loses valuable information about the state of the organization. Conflicts would become invisible – and therefore irresolvable.

The Semantic Graph Layer must therefore do more than pure exposition.

6. Architectural principle

Internal Multiplicity, External Determinism

The Semantic Graph Layer operates according to a dual truth model:

Internal level (Graph Core): multiple, conflict-capable
External level (Machine Interface): deterministic, collapsed

Contradictions are not avoided, but structurally recorded. However, they are not exposed unfiltered.

7. Assertion model

The Semantic Graph Layer models relations as assertions. An assertion is a verifiable claim about an entity in a defined context.

7.1 Scoped Assertions (Context Dependence)

Many apparent contradictions are contextual differences. Different values can exist in parallel, provided their scope is clearly modeled (e.g., region, organizational unit, market).

The graph holds these assertions in parallel.
The exposition is context-dependent.

7.2 Provenance & Confidence

Real contradictions are stored as competing assertions, not overwritten.

Each assertion has:

a source (Provenance)
a weighting or confidence

The graph does not evaluate “truth”, but evidence.

7.3 Temporal Validity

Statements have temporal scopes. What was true yesterday may be invalid today without having been false.

The graph allows parallel assertions along the timeline without logical collision.

8. Conflict resolution

Conflict resolution is not an ad-hoc process, but part of the architecture.

The Semantic Graph Layer distinguishes three phases:

Ingestion: Recording competing assertions, marking conflicts.
Resolution: Application of governance rules, confidence logic, or temporal prioritization.
Exposition: Output of exactly one canonical assertion per context to external systems.

This canonical assertion forms the Collapsed State.

9. Relationship to other layers

Entity Inventory: defines identity and existence.
Semantic Graph Layer: models meaning, relation, and validity.
Transport-Safe Content Layer: reflects selected, canonical graph truths in a retrieval-resilient manner.
Projection / Machine Interface Layer: exposes exclusively deterministic states.

Summary

Meaning that exists only in the text is unstable for machines. Meaning that is only implicit is not verifiable. The Semantic Graph Layer shifts meaning from interpretation to architecture. It allows internal polyphony without creating external ambiguity.

Internal Multiplicity, External Determinism is not a compromise, but the prerequisite for truth to remain stable under probabilistic systems.

Link tip

Semantic Graph Engineering: technical rules, data models, and validation criteria for the construction and operation of the Semantic Graph Layers in Aivis-OS

Architecture Overview

All Aivis-OS Core Architecture

Cluster-Level Entity Inventory Strategy