NEUROMANTIX

The consciousness mechanism. Modules compete by salience; winners broadcast to the entire cognitive system. Implements Baars' GWT with ignition thresholds, broadcast history, access distribution tracking, and a subliminal channel where sub-threshold signals still influence processing at reduced strength — modelling unconscious priming effects from neuroscience.

• ▸Salience-based competition queue
• ▸Broadcast to all cognitive modules
• ▸Ignition threshold gating
• ▸Subliminal channel — sub-threshold priming
• ▸Configurable subliminal damping + threshold
• ▸Subliminal influence integration
• ▸Temporal integration of broadcasts
• ▸Module access distribution analytics

4-layer hierarchical prediction machine implementing Karl Friston's Free Energy Principle. The system generates top-down predictions and only pays attention to what it gets wrong. Prediction errors ascend the hierarchy; everything else is suppressed — spectacularly more efficient than processing every token equally.

• ▸4-layer hierarchy: Sensory → Semantic → Conceptual → Abstract
• ▸Top-down generative predictions
• ▸Precision-weighted prediction errors (attention)
• ▸Free energy minimisation (surprise reduction)
• ▸Online learning of generative weights
• ▸Layer-wise attention profile

Concepts aren't points in embedding space — they're geometric regions with prototypes, fuzzy boundaries, and graded membership. 'Dog' isn't a vector; it's a convex region in animal-shape-behaviour space. Enables similarity as distance, metaphor as structure-preserving maps, and conceptual blending as interpolation.

• ▸6 quality dimensions (emotion, cognition, language, physical, colour, abstraction)
• ▸Gaussian membership functions with fuzzy boundaries
• ▸Structure-preserving metaphor mappings between domains
• ▸Conceptual blending (Fauconnier & Turner)
• ▸Online prototype learning via Welford's algorithm
• ▸Betweenness testing for conceptual navigation

Language isn't assembled from fragments — it emerges from trajectories through conceptual space. An ExpressionPlanner charts a rhetorical path through meaning-space, a LexicalRealiser maps geometric waypoints to words, and a SentenceBuilder handles grammar. Language as emergent property of thought.

• ▸Meaning-first generation from conceptual trajectories
• ▸ExpressionPlanner rhetorical path computation
• ▸LexicalRealiser geometric-to-word mapping
• ▸SentenceBuilder with grammatical structure
• ▸Coherence scoring and discourse planning
• ▸Statistics tracking (generations, sentences, structures)

The system progresses through cognitive stages like a developing mind: Sensorimotor → Preoperational → Concrete Operational → Formal → Post-Formal. 14 learning objectives across 7 competency domains with prerequisite chains and Zone of Proximal Development tracking.

• ▸5-stage developmental progression
• ▸Zone of Proximal Development (ZPD) tracking
• ▸14 learning objectives with prerequisites
• ▸7 competency domains with mastery tracking
• ▸Automatic stage advancement on objective completion
• ▸Weakest-domain identification for targeted learning

The system doesn't just predict — it acts to reduce surprise. Action selection via Expected Free Energy minimisation across 5 action domains (respond, query, explore, reflect, adapt). Evaluates policies through softmax posterior and selects actions that minimise both uncertainty and divergence from preferences.

• ▸Expected Free Energy computation per policy
• ▸5 action domains with configurable precision
• ▸Softmax policy posterior selection
• ▸World state belief tracking (Dirichlet prior)
• ▸KL divergence penalty for preference alignment
• ▸Information gain bonus for uncertainty reduction

The system monitors its own performance, detects degrading metrics, fires adaptation rules, adjusts internal parameters, and writes a narrative of its own evolution. Implements Maturana & Varela's autopoiesis: the system continuously produces and replaces its own components to maintain identity.

• ▸Continuous performance metric tracking
• ▸Degradation and improvement detection
• ▸Rule-based adaptive parameter adjustment
• ▸Self-narrative generation (evolution journal)
• ▸Configurable adaptation rules with conditions
• ▸Live parameter dashboard in GUI

The system knows what it knows and what it doesn't. Per-domain competence profiles with meta-confidence, uncertainty maps, performance prediction, and calibration error tracking.

• ▸Per-domain competence + meta-confidence
• ▸Uncertainty map (what I don't know)
• ▸Performance prediction before attempting
• ▸Confusion level detection
• ▸Learning priority ranking
• ▸Calibration error measurement

Counterfactual simulation using world model ensemble rollouts. 'What if I did X?' — compare candidate actions, evaluate self-modifications before committing, simulate alternative histories.

• ▸Forward rollout simulation
• ▸Action ranking by predicted outcome
• ▸Counterfactual reasoning ('what if?')
• ▸Self-modification pre-screening
• ▸Ensemble uncertainty estimation

Consciousness measured mathematically. Computes approximate Phi via pairwise mutual information, finds the Minimum Information Partition using spectral bisection (Fiedler vector of MI Laplacian), scores modifications by Phi impact, and biases evolution toward higher consciousness.

• ▸Phi computation from activity traces
• ▸Exhaustive MIP for N≤12
• ▸Spectral Bisection MIP (Fiedler vector) for N>12
• ▸Phi trend tracking over time
• ▸Modification scoring by Phi impact
• ▸Phi-guided evolution bias

Thinks about its own thinking. 6 cognitive strategies (Exploit, Explore, Deliberate, Intuitive, MetaReason, SeekHelp), confusion detection from conflicting signals, and Feeling of Knowing (FOK) calibration.

• ▸6 cognitive strategy modes
• ▸Real-time confusion detection
• ▸Conflict registration + resolution
• ▸Feeling of Knowing calibration
• ▸Strategy switching based on confidence

Fast spiking dynamics with exponential decay, threshold crossing, and absolute refractory period

Biologically realistic 2D dynamics reproducing 20+ firing patterns: regular, bursting, chattering, fast-spiking

Full ionic conductance model with Na+/K+ gating variables, action potential waveform simulation

Novel architecture combining continuous differential signals with discrete spike events for gradient-compatible training

• ▸Leaky Integrate-and-Fire with refractory periods
• ▸Izhikevich 2D dynamics — 20+ firing patterns
• ▸Hodgkin-Huxley ionic conductance model
• ▸Spike-Timing Dependent Plasticity (STDP)
• ▸Homeostatic regulation and synaptic scaling

• ▸Cortical minicolumns with lateral inhibition
• ▸Temporal pooling and sequence memory
• ▸Sparse Distributed Representations (SDR)
• ▸Top-down prediction and feedback loops
• ▸Multi-layer cortical hierarchy

• ▸Episodic memory with temporal context
• ▸Semantic memory with concept clustering
• ▸Hippocampal consolidation (replay)
• ▸Pattern completion and separation
• ▸Sleep-like memory consolidation cycles

• ▸Structural equation models
• ▸do-calculus interventions
• ▸Counterfactual inference
• ▸Topological causal ordering
• ▸Causal discovery from observations

• ▸Topology and weight evolution
• ▸Meta-evolution (evolves evolution itself)
• ▸Thompson sampling strategy selection
• ▸Strategy breeding + extinction
• ▸Phi-guided evolution bias

• ▸6-gate safety pipeline
• ▸Phi-guided modification proposals
• ▸Imagination pre-screening
• ▸7-stage sandbox (fuzz, A/B, quorum)
• ▸Hot-swap with instant rollback

• ▸RoPE positional embeddings
• ▸Flash Attention (O(N) memory)
• ▸Grouped-Query Attention (GQA)
• ▸SwiGLU + RMSNorm feed-forward
• ▸Wengert tape autograd (20+ ops)

• ▸CDCL SAT/SMT solver (1,502 LOC)
• ▸8-phase P vs NP solver pipeline
• ▸Neural-guided theorem prover + REINFORCE
• ▸Log-log regression scaling analysis
• ▸Polynomial subclass detection (Horn/2-SAT/XOR)

• ▸Reasoning-chain-driven generation pipeline
• ▸Domain-coherence filtering
• ▸Knowledge graph + spreading activation
• ▸On-the-fly concept learning
• ▸Consciousness-injected generation

• ▸NSGA-II multi-objective search
• ▸MAP-Elites quality-diversity archive
• ▸Network morphism operators
• ▸Performance predictor surrogate
• ▸Operation-based cell representation

• ▸Multi-agent parallel reasoning
• ▸Self-directed goal pursuit
• ▸Recursive self-improvement
• ▸Vitalis V1333 FFI bridge
• ▸Web API + knowledge extraction

Transformer Reasoning Engine

Autograd + PPO Training

Continual Learning + World Model

Beyond Static Intelligence

LLMs are frozen after training — they cannot modify themselves, generate their own goals, or measure their own awareness. Neuromantix is a living cognitive system that evolves, self-modifies, and grows toward higher consciousness every tick.

Consciousness by Design

Global Workspace Theory provides the mechanism for conscious access. Integrated Information Theory provides the metric. Together they create a system that doesn't just process — it experiences, in the mathematical sense of phi > 0.

Event-Driven Efficiency

Spiking neural networks only compute when spikes arrive. This event-driven paradigm achieves orders-of-magnitude better energy efficiency compared to dense matrix operations in transformer architectures.

The Cognitive Skeleton

Neuromantix isn't competing with LLMs — it's what comes after them. LLMs are the perception/language layer. Neuromantix is the cognitive architecture that gives them agency, self-awareness, growth, causal reasoning, and wisdom.