Integrated Architecture — SCC × ONN as one cognitive-reasoning system

The ultimate target of the entire programme — the reason every other page on this site exists.

The claim#

Perception, representation, reasoning, and control are one continuous mathematical object. The claim of this research is that Soft Cognitive Cohesion (SCC) and Ontology Neural Networks (ONN) — together with their mathematical substrate (RelationWorld Theory) and their control-theoretic closure (ORTSF) — can be composed into a single cognitive-reasoning architecture that carries structure without loss from the pre-objective soft field all the way down to an embodied control signal.

This is the design target. The rest is bookkeeping.

The four layers#

Layer 1 — Soft Cognitive Cohesion (pre-objective)#

Primitive. A soft cohesion field $u_t : X_t \to [0, 1]$ over a relational support space; the value $u_t(x)$ is the degree to which site $x$ participates in a cohesive formation.

Operators. Closure $\mathrm{Cl}_t$ (self-completion), distinction $\mathbf{D}_t$ (self-contrast), temporal transport $\mathbf{M}_{t \to s}$.

Energy. $E = \lambda_{\mathrm{cl}} E_{\mathrm{cl}} + \lambda_{\mathrm{sep}} E_{\mathrm{sep}} + \lambda_{\mathrm{bd}} E_{\mathrm{bd}} + \lambda_{\mathrm{tr}} E_{\mathrm{tr}}$ on the volume-constrained simplex $\Sigma_m$.

Output. A proto-cohesion diagnostic vector $\mathbf{d} = (\mathrm{Bind}, \mathrm{Sep}, \mathrm{Inside}, \mathrm{Persist}) \in [0, 1]^4$.

Ontological commitment. The field is primitive; objects are derivative. Deeper dive →

Layer 2 — RelationWorld (structural discovery)#

Substrate. A weighted relational field $\mathcal{W}_t = (V, W_t, g_t)$ where every edge carries both a positive weight and a group-valued transit.

Discovery. Fruits — low-conductance clusters detectable via the Cheeger condition $\phi_t(F) \le \theta$ — are the natural stable structures in the relational field. Doors are the boundary-adjacent singularities that signal contact with an exterior not explicitly modelled.

Output. A fruit set $\mathfrak{F}_t$ together with its existence triple $([A_\infty], \Sigma, \mathbf{e})$ — a gauge-invariant portrait.

Role in the architecture. This is the layer that makes SCC concrete: the graded soft field acquires a relational substrate with a theory of what counts as existing. Deeper dive →

Layer 3 — Ontology Neural Networks (relational semantics)#

Substrate. A fruit-node semantic graph — fruits as nodes, their inter-fruit relations as edges, the ontology's type structure as type algebra on both.

Process. Constraint-based semantic inference on this graph, projecting onto a topology-aware manifold. The ONN's internal state is not a flat vector but a typed object carrying the structure of the target ontology.

Output. A semantic structure augmented with action predicates — what follows from what the fruits mean.

Role in the architecture. ONN is what turns discovered structure into actionable meaning, with invariants that can be read as cohomology classes. Deeper dive →

Layer 4 — ORTSF (embodied action)#

Substrate. A delay-robust control fabric consuming ONN semantic output.

Process. Meaning-preserving trajectory synthesis under stochastic communication, sensing, and compute delay.

Output. A real-time embodied control signal with a delay-robust stability margin — a standard plant-pole phase margin, not a cohomological bound (see the ONN audit).

Role in the architecture. This is what closes the loop from cognition back to the world. Deeper dive →

Why each layer requires the next#

Layer 1 — SCC
  "Objects are given, but how does anything first hold together?"
  → Soft field, four energy terms, graded cohesion.
      │ "u_t is abstract. What is the relational substrate?"
      ▼
Layer 2 — RelationWorld
  "If cohesion is relational, what makes relations cohere?"
  → Weighted gauged graph, fruits, doors, existence.
      │ "Fruits are discovered. How do they interact semantically?"
      ▼
Layer 3 — ONN
  "Fruits are topological units — how do we reason between them
   while preserving their internal structure?"
  → Topology-aware constraint inference, type-preserving projection.
      │ "Semantics are inferred. How do we embody them in action?"
      ▼
Layer 4 — ORTSF
  "Semantic structure is defined. How do we preserve that meaning
   in real-time control under stochastic delay?"
  → Delay-robust control via a standard plant-pole delay margin.

Forward flow and backward flow#

Forward flow is the bottom-up emergence of meaning:

Backward flow is the top-down validation and refinement:

The architecture is only whole when both flows are closed — the system not only produces behaviour but uses the outcomes of that behaviour to update its beliefs about what exists, what is, and what it should do next.

Why this integration matters#

Most architectures for perception-to-control keep representation learning, semantic reasoning, and control synthesis as separate problems with separate guarantees. The integrated architecture proposed here insists that the three share a single mathematical substrate — topology — and that guarantees in one layer translate directly into guarantees in the others.

This was the programme's organising hope, but on the ONN + ORTSF side it has since been audited and does not hold in its strong form: the control margin is a standard delay margin driven by the plant pole and is decoupled from — not governed by — the learned ontology's cohomology, and measured higher-order structure adds no information beyond pairwise (the Two Ceilings boundary). The SCC / RelationWorld theorems of Parts I–II stand on their own; the "single certificate" coupling them through cohomology is not established.

Where this stands today#

• Layer 1 (SCC) — 68 Cat A theorems (98 total claims, ~70% proved) at CV-1.17 (sealed 2026-05-15), 3 active HIGH OPs (OP-0005 K-Selection, OP-0008 σ^A K-jump non-determinism, OP-0009 Multi-Formation Foundations); OP-0006 boundary precision was resolved at W6 (T-OP6-B). The W4-original Critical OPs F-1 / M-1 / MO-1 were resolved/clarified/sidestepped on 2026-04-24. Layer 1 includes the σ-framework (Commitment 14 multi-static σ + Commitment 16 two-tier K_field/K_act decomposition), the first multi-formation Cat A theorem T-L1-F (CV-1.5.2, Hard-Bar / Active-Count Bridge under the L1-J regime on shared-pool architecture I9′), the W6 P-F-A1 Package I stochastic foundation (fully Cat A), and the W7 additions T-Temporal-Identity (a) (Cat A, CV-1.12, via H-SINK) and T-CC-StableK-Kernel (Cat B, CV-1.17, H-COMP-KERNEL closed). See the status page.
• Layer 2 (RelationWorld) — Theorems A–H proved. See Part I and Part II summary.
• Layer 3 (ONN) — The framework paper is published, but its central higher-order thesis was subsequently audited to a scoped No-Go boundary (measured higher-order/cohomological structure adds no information beyond pairwise). See the ONN status; live work moved to ULR.
• Layer 4 (ORTSF) — Reduces to a standard, verified delay-margin certificate ($\Delta t_{\max} = \varphi_{\mathrm{PM}}/\omega_c$, plus a sufficient small-gain condition); the earlier cohomological-Lyapunov and τ_max = 177 μs claims are withdrawn.
• Integration theorem — in progress. The piece that turns four separate results into one architecture.