Architecture

System Architecture

Oneiros Engine adopts a state-based, closed-loop architecture for reasoning under uncertainty. Decisions are modeled as evolving states rather than linear pipelines or single-shot optimizations.

The architecture emphasizes traceability, revision, and human-in-the-loop judgment—principles derived from long-term practice in complex workflows and aligned with research and STEM learning needs.

State-based Loop Traceability Revisable Equilibrium Human-in-the-loop

1. Architecture at a glance

At a high level, Oneiros Engine operates as a closed-loop system. Information is perceived and framed, structured inquiry is conducted under constraints, a provisional equilibrium is reached, and the resulting rationale is preserved in memory—ready to be revisited when context or evidence changes.

Perception & Framing → attend to and weight incoming signals

↓

Inquiry → organize evidence, compare hypotheses, manage trade-offs

↓

Decision Equilibrium → reach a revisable, temporary balance

↓

Memory → preserve rationale and state snapshot

↓

Perception → re-enter loop when context updates

2. Core loop: Perception → Inquiry → Decision Equilibrium → Memory

The core loop formalizes how decisions evolve over time. Perception is treated as selective framing rather than raw input. Inquiry captures structured reasoning under constraints. Decision equilibrium represents a stable but revisable stance, and memory maintains continuity without freezing conclusions.

This loop reflects practical constraints observed in real workflows: delayed information, partial evidence, team handoffs, and changing objectives.

State Model

3. State & trace model

Each state in Oneiros Engine can be represented by a compact set of attributes:

Evidence: selected and weighted observations
Assumptions: provisional premises subject to decay
Constraints: costs, risks, policies, or resource limits
Rationale: why a balance was reached at this point

By preserving this trace, the system supports revision without erasing history—an essential feature for accountability, learning, and collaborative work.

Interface Layer

4. STEM-aligned interface layer

The architecture exposes an interface layer that translates internal states into reflective prompts. These prompts act as triggers for reasoning, not as answers. They align with STEM practices such as systems thinking, evidence-based claims, reasoning under uncertainty, and iterative revision.

Uncertainty Trigger

Highlights missing or ambiguous evidence

Conflict Trigger

Surfaces competing constraints or interpretations

Revision Trigger

Invites controlled updates without full reset

Narrative Layer

5. Narrative consolidation

The final architectural output may take the form of a narrative artifact: a concise account of how a decision evolved across states. Narrative here serves consolidation and communication, allowing reasoning to be reviewed, transferred, and critiqued without framing earlier states as errors.

6. Scope & boundaries

Oneiros Engine does not prescribe domain-specific solutions, simulate biological mechanisms, or claim therapeutic effects. It provides an abstract, research-oriented architecture for modeling and teaching decision-making in complex environments.

For conceptual framing, see Concept. For usage contexts, see Applications.