Frequency Architecture

14 Neural Frequency Bands

Each of SOV3's 14 neural models is tuned to a specific frequency band. Together, they create a multi-spectral picture of reality — from millisecond sensor pulses to month-long strategic patterns.

Cross-Band Binding Example: ISR Pipeline

ISR Pipeline — 14-Band Frequency Decomposition:

υ (100-500Hz):  Drone camera frame arrives. Orion detects movement.
δ (0.5-4Hz):    Moondream identifies object class (vehicle, person, UAV).
χ (3-7Hz):      Kimi fuses image + ADS-B telemetry + terrain.
β (12-30Hz):    Falcon3 runs threat classification code.
γ (30-100Hz):   Qwen routes to appropriate expert. Binds detection to context.
θ (4-8Hz):      Llama checks historical patterns. "This route seen before?"
α (8-12Hz):     DeepSeek reasons about intent. "Supply convoy or attack formation?"
μ (8-13Hz):     Nomic finds semantically similar past incidents.
φ (1-3Hz):      Nemotron Nano checks JSP 936. "Is response proportional?"
σ (12-14Hz):    Nemotron 30B assesses care. "Civilian proximity risk?"
κ (0.1-1Hz):    Mamba-2 checks long-term patterns. "This is new behaviour."
ρ (0.01-0.1Hz): MoE decides which 8 experts to activate.
τ (0.001-0.01Hz): Quantum VQE optimises resource allocation.
ψ (0.05-0.5Hz): BFT Council votes. "ENGAGE or HOLD?"

Total time: 8-12 seconds. Traditional: 45 minutes.

Mamba-2 State-Space Duality (SSD)

Mamba-2 is SOV3's long-memory engine. Unlike Transformers (which attend to all past tokens — O(n²) cost), Mamba-2 compresses the entire history into a 16-dimensional state vector. This state vector is the "vibe" — the frequency-domain representation of everything SOV3 has experienced.

State Vector Dimensions

Mamba-2 vs Transformer

# Mamba-2 state update (simplified)
# Every new observation updates the 16-dim state vector

def mamba2_update(state, observation, dt):
    """SSD: State-Space Duality update."""
    A = state_matrix(state)        # 16×16 transition matrix
    B = input_matrix(observation)  # 16-dim input projection
    C = output_matrix(state)       # 16-dim output projection

    # Continuous-time state update
    d_state = A @ state + B @ observation

    # Discretize with zero-order hold (ZOH)
    state_new = state + d_state * dt

    # Output = state projected through C
    output = C @ state_new

    return state_new, output

Stochastic Resonance — When Noise Helps

Stochastic resonance is a counterintuitive phenomenon: adding optimal noise to a weak signal amplifies it rather than drowning it. SOV3 uses this principle to detect threats that would be invisible to conventional filtering approaches.

How It Works

Resonance Profile — Optimal Noise per Feature

Quantum QAOA Optimisation

The optimal noise level (σ) for each feature is computed by a variational quantum algorithm (QAOA — Quantum Approximate Optimisation Algorithm) running on SOV3's quantum batch. The QAOA finds the σ that maximises the probability of the signal crossing the detection threshold — a non-convex optimisation problem that classical methods struggle with.

# QAOA care + resonance optimisation (simplified)
# Runs on M2 Mac nightly, pushes results to SOV3 memory

qaoa_result = mcp_call("sov3-federation", "run_quantum_batch", {})
# Returns: optimal σ for each feature
# Pushes to SOV3 memory for real-time use

The Intuition Engine

The Intuition Engine is SOV3's highest-level cognitive output. It ingests SIGILs at 1Hz from all 30 MCP servers, 14 neural models, and 33 council agents — and produces "intuition": a 16-dim Mamba state + energy + complexity + a wisdom string. This is what makes DEFONEOS feel like it "knows" something before the data confirms it.

Architecture

┌──────────────────────────────────────────────────────────┐
│                 SOV3 INTUITION ENGINE                     │
│                                                           │
│  30 MCPs ──┐                                             │
│  14 Models ─┤                                             │
│  33 Council─┼──→ SIGIL Bus (1Hz) ──→ Mamba-2 SSD ──→     │
│  Vault     ─┤         ↑                    ↓              │
│  Memory    ─┘     Stochastic         16-dim State         │
│                    Resonance        ┌──────────┐          │
│                    Injection        │ S₁-S₁₆   │          │
│                                     └──────────┘          │
│                                          ↓                │
│                                    Wisdom Emission        │
│                                    (natural language)     │
│                                          ↓                │
│                                    BFT Council Review     │
│                                    (33 agents vote)       │
│                                          ↓                │
│                                    SIGIL Chain            │
│                                    (Ed25519-signed)       │
└──────────────────────────────────────────────────────────┘

Intuition History — Last 24 Hours

Lead Time Advantage

The Intuition Engine provides 20-60 seconds of lead time over traditional alerting — it detects patterns before they cross formal thresholds. In a counter-drone scenario, 20 seconds of lead time is the difference between detection and interception. In cyber defence, it's the difference between containment and breach.

MCP Tools — Frequency & Intuition