# Synthetic Cortex — full site text > Plain-text copy of every page on https://syntheticcortex.dev, generated 2026-11-15. Canonical pages are linked under each heading. --- # Synthetic Cortex — Cognitive Architecture for Language Models URL: https://syntheticcortex.dev/ Synthetic Cortex is an external cognitive architecture for large language models: a 13-channel synthetic neurochemical state, threshold-gated parallel layers, associative memory and context synthesis. Gen 1: L1, L2 and M1 (L1.5) shipped. Synthetic Cortex · Gen 1 · Cognitive architecture ## Thinking,re-architected. Synthetic Cortex is an external cognitive architecture that wraps a language model and decides how it thinks: which memories surface, which distant concepts are pulled together, when to reason step by step and when to stop. Not a bigger model. A better-organised one. Explore the architecture → See it running in Generation Network ↗ L1 · shippedL2 · shippedM1 (L1.5) · shipped on mobileGen 1 · in active development Scroll 00 — The system ### Signal path, end to end. Every input crosses this map before a language model writes a single word. Tap any node. Lime lines carry the main signal, cyan lines carry emotional state. Red nodes were closed by the agent that prepared this page; dashed nodes are branches not yet disclosed. Signal flowEmotional stateRedacted by agentUndisclosed branchScroll sideways on small screens → Explore the details →The seven research phasesEvery module, mapped to the brain Pipeline node× #### Node Go toArchitectureLimbic-2Cortex AtlasNeurochemistry 01 — The pipeline ### One input. Dozens of parallel processes. One denser context. A conventional chat model receives text and predicts text. Everything it can use in a given moment has to fit inside its context window, and what gets into that window is usually decided by recency or keyword similarity. Synthetic Cortex inserts a regulatory pipeline between the input and the model. Stage 01 #### Affect scoring Every input — a message, a letter, a behavioural signal — is scored on three continuous axes: valence, arousal and dominance. No emotion labels, only coordinates. Stage 02 #### Neurochemical state Those coordinates drive a 13-channel synthetic neurochemical state. Channels push and suppress each other, carry momentum from earlier turns, and are pulled back toward balance by a homeostatic regulator. Stage 03 #### Threshold gates The state acts as a field of thresholds. Crossing one fires a process: an associative sweep, a memory consolidation cycle, a selective-perception pass, a self-review, a deep pattern inference. Stage 04 #### Context synthesis What the processes return is compressed into hidden context, the model's sampling behaviour is tuned to the current state, and the output re-enters the pipeline as a new stimulus. The loop never really stops. Inputs shift the state; the state shapes the reply; the reply shifts the state again, and that shapes the next input. Between turns, background processes keep working: records are broken into fragments, linked to each other, consolidated or forgotten, and re-weighted by how emotionally intense they were when they arrived. What reaches the model at the end is not a longer prompt. It is a more selective one. See the full flow, including the parts we keep closed, on the architecture diagram. 02 — What it changes ### From statistical recall to regulated thought. Synthetic Cortex does not feel. It is a mathematical model of what the brain does while feeling: the filtering, prioritising and gating that emotion performs on cognition. The aim is the same kind of outcome the brain reaches — fast, contextual, occasionally surprising decisions — reached through a different substrate. Emotions narrow the options. Logic justifies the choice. Associative intelligence #### It retrieves what is related, not only what is near. Similarity search returns the records closest to a query. Associative recall also brings in records that belong to the same broader area — close in meaning but not directly on topic — whenever there is room in the context. Spreading activation across a graph of fragments lets one memory call up others. Expected outcomeAnswers that connect topics you never placed side by side. Emotional intelligence, computed #### Priority is decided by state, not frequency. The neurochemical state determines what is salient and what fades. Among similar records, an emotional recall ratio works alongside semantic similarity, so the most emotionally significant memory near a topic comes forward first — and a reactive state processes the same input more strongly than a calm one. Expected outcomeResponses that weigh what matters in this context, not what is statistically common. Context-window economics #### Same window, more signal. Output quality depends on how much meaningful information fits in the context window. Instead of padding the prompt, the cortex pre-selects and densifies: concepts that sit far apart in the model's geometric meaning space are brought into the same window through associative context management. Pre-computed emotional baselines mean the system rarely recalculates from scratch. Expected outcomeFewer tokens wasted on raw text, more of them spent on relationships. Controlled creativity #### Two noisy processes, one coherent output. An affective process and a logical process are each capable of irrational deviations on their own. Integrated, they constrain each other: associative sweeps widen the field, homeostasis and logic narrow it again. Creative expansion fires only when thresholds are crossed — neither constant randomness nor rigid determinism. Expected outcomeUnexpected connections that still hold together. Surfacing latent knowledge #### A model knows more than it says. Knowledge inside a language model lives in dense "token ecosystems" that are reachable only along certain paths. Ordinary prompting keeps hitting the statistically dominant ones. By changing the path — what enters the context, and from which associative and emotional angle — the cortex reaches regions of the model's own knowledge that normal prompting leaves dormant. Expected outcomeRare patterns in the training data become reachable again. Time as an ingredient #### It learns slowly, on purpose. Information moves through layered memory — short-term, transitional, long-term — is consolidated when it proves useful or emotionally strong, and is forgotten under a homeostatic budget, leaving a trace behind. The architecture trades instant recall for depth of integration. Expected outcomeThe longer it runs, the further its outputs move away from a raw model's. 03 — Scale of the cortex ### Not one network. An ecology of them. Each module can be developed, replaced or switched off on its own, yet every module listens to the same neurochemical state. That shared state is what turns a collection of components into one system. Neurochemistry #### 13 coupled hormone and neurotransmitter channels Spiking networks #### 10 spiking neural networks — three core, seven hybrid Inference hierarchy #### 14 stacked layers across Equalia and the deep layers Parallel spaces #### 2 manifolds — one for meaning, one for emotion Memory tiers #### 3 short-term, transitional, long-term — each with forgetting Hidden context #### 12+ channels feeding the model what it was never explicitly told 04 — The family ### Synthetic Cortex, Gen 1. Three builds are complete. They share the same idea and differ in how deeply they can reach into a model. L1Gen 1 ShippedThe first and simplest layer. Synthesises a neurochemical state and uses its weights to trigger processes — tool calling, hidden-context injection and module activation. L2Gen 1 ShippedThe main system and the most comprehensive build. Attaches to open models and works inside inference: latent-space manipulation and activation steering, driven by weights read from the model itself. M1= L1.5 Shipped · mobileThe mobile cortex. Designated L1.5, it is in practice the leanest build in the family: many features were removed and others re-engineered so the whole cortex runs locally on a phone. Small on-device models stand in for what L2 does inside a model. Status #### Gen 1 is still being built. L1, L2 and M1 are the first three milestones, not the finish line. Research tracks for internal learning, manifold geometry, default-mode dynamics and cultural resonance continue inside Gen 1. Gen 2 is planned. L1 ✓L2 ✓M1 ✓Phases 2 · 5 · 6 · 7 →Gen 2 Open the roadmap → 05 — First product ### Generation Network. The first product built on Synthetic Cortex. Generation Network runs the M1 mobile cortex on the phone itself: your Reflection — a digital copy built from your letters, chats and in-app behaviour — synthesises its own neurochemical state, consolidates memories while the phone is idle, and learns to resemble you over weeks rather than minutes. Personal mental models stay on the device by default. That constraint shaped the whole M1 build: everything had to be light enough to run locally. Get it on Google Play ↗ generationnetwork.app ↗ What the cortex does inside it → REFLECTION · M1mood drifting toward calm-positiveconsolidation queued · idle 06 — Go deeper ### Choose a depth. The site is layered the way the system is. Start anywhere. Technical #### Architecture → The full pipeline diagram, the seven research phases, Manifold A and B, implicit relation engineering, activation-as-reasoning. Modules #### Cortex Atlas → Every module in the mobile cortex, mapped to the brain process it imitates and the outcome it is built to produce. Interactive #### Neurochemistry → The 13-channel interaction matrix and a live emotion-frequency lab you can push out of balance. Timeline #### Roadmap → From Limbic v1 to Gen 1 today, and what comes next. Plain language #### FAQ → Synthetic Cortex explained without jargon. Writing #### Blog ↗ Essays and updates, published on the Generation Network blog. --- # Architecture — Synthetic Cortex Pipeline, Phases and Parallel Spaces URL: https://syntheticcortex.dev/architecture/ Technical overview of the Synthetic Cortex architecture: the full pipeline diagram, threshold-gated parallel layers, the seven research phases, Manifold A and B, implicit relation engineering, activation-as-reasoning and the M1 mobile cortex. Architecture · Technical ## The cortex, mapped. Synthetic Cortex defines how thinking is structured, not what a model is trained on. It attaches to an existing language model and organises inference through parallel, specialised subsystems that all read from one shared neurochemical state. This page shows the pipeline as far as we publish it. Open the diagram ↓The seven phasesHow M1 differs 01 — Pipeline ### Signal path, end to end. Tap any node. Solid lime lines carry the main signal, cyan lines carry emotional state, grey lines are internal hand-offs. Red nodes were closed by the agent that prepared this page. Dashed nodes are branches we have not disclosed yet. Signal flowEmotional stateRedacted by agentUndisclosed branchScroll sideways on small screens → The pipeline in words - Input streams arrive: conversation, letters and messages, behaviour, device state. - A relay stage routes them and pre-filters by salience. - Affect scoring places each input on valence, arousal and dominance. - The Limbic-2 engine updates thirteen coupled neurochemical channels; a mood space, an emotional trace memory, a hypothalamic body signal and a homeostatic regulator act on it. - Threshold gates translate the state into which processes fire. - Parallel layers run: spiking networks, layered memory, Manifold A ⇄ B, the perception network, the default mode network, associative recall, selective perception, self-space, metacognitive review, and one withheld layer. - Equalia and the deep layers stack inference seven levels deep each. - Hidden context synthesis gathers every channel into what the model sees; output steering and expert delegation feed it; sampling coupling is withheld. - The language model answers. The answer re-enters at affect scoring. What is not shown #### Closed on purpose. Gate definitions, the conversion from affective coordinates to neurochemical levels, the coupling between state and generation parameters, the exact channel set of the hidden-context manager, and several modules in development are not published. The diagram is accurate in shape and deliberately incomplete in detail. WhyThe value of the architecture is in its orchestration. The components are public science; the way they are wired together is not. 02 — How it differs ### Not a prompt technique. Not a RAG pipeline. Not fine-tuning. Most systems change what a model sees or what it has memorised. Synthetic Cortex changes the conditions under which it thinks. | | Dimension | Prompted LLM | RAG pipeline | Fine-tuned model | Synthetic Cortex | Decides what the model sees | The prompt | Similarity to the query | Already in the weights | A neurochemical state, gates and associative recall | Reasoning starts | When asked | When asked | When asked | When an internal gate crosses its threshold | New knowledge lives | In the context window | In an external index | In changed weights | In layered memory, parallel spaces and hidden context (L2: latent space) | Base weights | Frozen | Frozen | Modified | Frozen | Retrieval timing | — | Once, before inference | — | Re-evaluated as inference evolves (L2) · emotion-weighted (M1) | Role of emotion | Tone, if any | None | Style learnt from data | Central regulator of priority, gating and consolidation | Forgetting | When the session ends | Rarely | Risk of catastrophic forgetting | Homeostatic, tier by tier, leaving traces | Explainability | Low | Cited sources | Low | Gates crossed, channels fired, per-module logs From tokens to meaning #### Pattern engineering, not prompt engineering. Prompt engineering works on what the model observes at the surface. The cortex works one level down: on how the model relates, organises and navigates meanings it has already inferred. Tokens are entry points into that space, not the objects of reasoning. Implicit relation engineering Versus classic data science #### Relations are designed, not just discovered. A conventional data pipeline engineers features and lets correlations emerge from volume. Implicit relation engineering shapes the relational geometry itself — which concepts resonate, which decay — so new information lands in an existing network of meaning instead of being learnt from scratch. Read the method Versus agent frameworks #### Orchestration by state, not by script. Agent frameworks route by rules and planners. Here the same input can fire different modules on different days, because routing depends on an evolving internal state with momentum, coupling and balance — closer to a regulated system than a flowchart. Activation as reasoning 03 — Research programme ### Seven phases, one cumulative architecture. Each phase addresses a specific limitation of current language models and adds a capability on top of the previous ones, without growing the core parameter count. Open any phase for its dossier. PHASE 01 #### Emotional reasoning via neurochemical modelling Neurochemical signals modelled mathematically and mapped onto continuous control variables that reshape inference. CompletedDOSSIER PHASE 02 #### Internal learning through dynamic adapters Knowledge embedded during interaction while the base model stays frozen. In researchDOSSIER PHASE 03 #### Latent space injection for capacity expansion Structured knowledge injected into latent space during inference, without scaling parameters. CompletedDOSSIER PHASE 04 #### Activation-based chain-of-thought extraction Reasoning drawn from the model's own activations instead of imposed prompt structures. Largely completeDOSSIER PHASE 05 #### Implicit relation engineering & manifold geometry Meaning as trajectories on a curved surface rather than distances in a flat space. In developmentDOSSIER PHASE 06 #### Associative reasoning via default-mode dynamics Controlled traversal of concept space beyond the immediate task. In developmentDOSSIER PHASE 07 #### Cultural resonance & contextual relativity The same concept interpreted through different cultural lenses. In developmentDOSSIER 04 — Limbic-2 ### Emotional loads as an inter-module protocol. Limbic-2 is the current emotional engine. Its predecessor, Limbic v1, was retired and archived; the present design goes beyond representing affect and doubles as a communication protocol between cortices. In the main system, parallel auxiliary computations run during inference across the later transformer layers — where abstract patterns have already formed — and the contextual signal extracted there is analysed on valence, arousal and dominance. The dominant context is amplified, alternative contexts are kept as controlled deviations, and the result is re-expressed as thirteen neurochemical values. Each cycle integrates with the previous one in a way that lets older cycles fade gradually rather than averaging them away, so the state has memory without being stuck. The output is a set of emotional loads that every other module is normalised against. Direct effects #### Inside the architecture. Emotional loads guide internal learning, latent injection and activation steering; they shape activation-based reasoning, default-mode association and cultural resonance. Weights produced by the model are deliberately modulated with these loads, triggering different probability protocols and shifting the overall reasoning flow. Indirect effects #### Thresholds for everything else. The same values set the thresholds for memory synchronisation, arousal regulation, salience filtering, episodic updates, contextual persistence and logical consistency — and for any external module a developer attaches. The system stays in balance because every operation is computed relative to its current state. Classified #### Affect → neurochemistry weighting The mapping that turns valence, arousal and dominance into thirteen channel levels, and the coupling between those levels and generation parameters, are withheld in this release. DA v·a·d ███ · 5HT ███ · OXT ███ · AVP ███ · END ███ · EPI ███ · NE ███ · CORT ███ · T ███ · E/P ███ · GABA ███ · GLU ███ · PRL ███ See the public interaction matrix and the live frequency lab → 05 — Parallel spaces ### Manifold A for meaning. Manifold B for emotion. Language models typically process information in a flat, Euclidean space where distances are straight lines and activation spreads globally: when one concept lights up, distant and irrelevant concepts are pulled into the computation too. A manifold organises information along a curved surface. Points that look far apart in vector space can be close along the surface, and what matters is proximity along it. The cortex keeps two such spaces in parallel. A state computed in one is not copied into the other as raw data; it crosses as an emotional context and a priority signal. Each point interacts only with its local neighbourhood, so the system reasons over a small relevant region instead of the whole space — which is also how the cost of manifold methods is kept down. Model weights stay fixed; only the path the data takes is curved. Manifold A #### Directed semantic re-mapping. A semantic core is extracted from internal activations and projected along directions that correspond to different conceptual or cultural frames, then mapped back to text. Abstractions emerge from relational convergence rather than from memorised sequences. Inside Manifold A Manifold B #### An isomorphic emotional twin. Same geometry, different job. Emotional loads from the current input, recent dialogue and episodic history are mapped onto every point of A. When accumulated load crosses a threshold, B can redirect A's reasoning trajectory. Inside Manifold B On the phone #### Manifold memory in M1. The mobile cortex keeps every record in two parallel spaces. When a topic comes up, emotionally intense records close to it come forward, and the Reflection takes them into account — with a ceiling on how strongly that shows. See it in the atlas 06 — Activation as reasoning ### Reasoning as a computation, not a narrative. Chain-of-thought prompting asks a model to write its reasoning out. That improves local coherence but stays at the level of text, and it remains tied to the distributions of the training data. In the main system, reasoning is grounded in the model's real-time activations: the cortex observes which semantic relations the model itself activates and raises their neighbouring clusters into the active context. Retrieval is not a single step before inference; targets are re-evaluated as decision-layer activations evolve, tying vector proximity and decision dynamics into one flow. In the emotional engine, the trigger for this mode is chemical: when the glutamate-analogous channel crosses its gate, chained reasoning engages; when the prolactin-analogous channel crosses its own, the loop is closed. No prompt tells the model to "think step by step". Benefits #### Faithful, controllable, debuggable. No weight updates, no fine-tuning; the architecture monitors existing representations and builds relational structure on top of them. Particularly suited to research-heavy domains and to agent reflection phases that need adaptive context reconstruction. Full dossier Open challenges #### Stated plainly. Activation-guided selection can drift into confirmation bias if early anchors over-constrain retrieval. Monitoring adds latency, mitigated by parallelism. Quality depends on the embedding pool behind it. 07 — Capacity & context ### More knowledge without more parameters. Latent injection · L2 #### Knowledge placed where the model thinks. Instead of storing information in weights or passing it as raw text, structured knowledge is injected into latent space during inference, where it interacts with internal representations as if it were already encoded. Core parameters remain unchanged, so cost and instability stay low. Phase 3 dossier Context density · M1 #### Hidden context on a phone. Without access to a model's internals, the mobile cortex arranges data on the device with small embedding models and injects it as hidden context. Concepts that sit far apart in the model's meaning space are brought into one window — the indirect route to the same effect. How M1 differs 08 — Creativity module ### Analytical rigour and associative reach, decoupled and re-coupled. Richer relational training makes models more creative and less consistent. Most systems settle on a compromise. The cortex separates the two: an analytically oriented layer produces structured reasoning and extracts conceptual anchors; a specialised associative layer, trained on deliberately dense cross-domain relations, expands only within those anchors; the analytical layer then re-grounds the result. Expansion mode is entered only when synthetic affective signals cross their thresholds. Emotion-gated #### Moments of depth without persistent instability. Because the creative layer fires on affective thresholds, enrichment appears spontaneously where context calls for it — not on every turn. Full dossier Restricted #### External MoE experts Completed. Information access restricted for commercial confidentiality. 09 — Cultural relativity ### Meaning is situated, not averaged. Training dynamics push culturally contingent concepts — legitimacy, respect, authority, family, freedom — toward one dominant statistical interpretation. The result is abstraction loss: the model stores an average where a society has a frame. The cortex targets the relational geometry around a concept, reshaping how activation propagates through its neighbours so that interpretation shifts with the active cultural resonance profile. Emotional loads keep the module dormant unless a dominant perturbation calls for it. In Generation Network, a first consumer-facing expression of this idea is the Time Machine: an idea is read through how different historical periods would receive it. Cultural relativity dossier + 10 — M1 · the mobile cortex ### L1.5: the leanest build, by necessity. The main system connects to open models and reads weights during inference to synthesise its neurochemical state; it can then steer the model's geometric latent space directly. A phone cannot carry that processing volume, and the mobile build was designed to keep personal mental models on the device. So M1 removes a large part of L2 and re-engineers the rest. - State without the model's weights. A predefined weighting route with an adapted dominance tone replaces weights read from inference. The evaluation system is similar; the advantages of model-derived weights are given up. - Steering without access to internals. Small external models embed data on the device and steer its arrangement locally; hidden contextual information is injected to shift the semantic space. Results are close, especially for semantic-space shifts, though not identical to L2. - Emotion-triggered recall. When certain emotional states are triggered, short inferences are retrieved from local memory, selected by an emotional recall ratio alongside cosine similarity. Close-but-unrelated records can join associatively, the way an unrelated memory surfaces mid-conversation. - Slow learning by design. Background learning runs whenever the phone is idle and yields to the user immediately. Dedicated components watch battery and memory. The outcome is not L2 on a phone. It is a system that reaches several of the same effects indirectly — and does far more than an average chat layer. Every M1 module is listed in the Cortex Atlas. 11 — Position ### What this is, and what it is not. It is #### A formal emulation of mechanisms. An attempt to abstract, mathematically, the functional processes that give rise to reasoning in biological cognition — learning, using what was learnt, adapting, solving problems — and to unify them under one architecture that is open in two directions: special-purpose "brain regions" for specific needs, and broader cognitive capability on the same infrastructure. It is not #### AGI, a mind, or a feeling machine. The project does not aim to build artificial general intelligence or an independent intelligent system. Its neurochemistry is a set of mathematical models. Benchmarking is ongoing; claims on this site describe design intent and internal observations. Pipeline node× #### Node Go toLimbic-2ManifoldsCortex AtlasNeurochemistry Phase 1 · Completed× #### Emotional reasoning via neurochemical modelling Human decision-making is not purely logical; it is continuously modulated by hormones and neurotransmitters. Phase 1 models these signals mathematically and maps them onto latent modulation coefficients that influence internal activations during inference. Rather than encoding emotions as symbolic labels, synthetic affective states are continuous control variables that reshape the latent geometry of the model. Emotional context therefore changes relevance weighting, associative strength and inference trajectories. The result is emotion-conditioned reasoning patterns — the first step toward synthetic affective cognition. Thirteen channels are modelled: dopamine, serotonin, oxytocin, vasopressin, endorphin, adrenaline, noradrenaline, cortisol, testosterone, estrogen/progesterone, GABA, glutamate and prolactin. They interact as coupled oscillators — directional, context-dependent and able to reverse as conditions change. In neuroscience this is called effective connectivity; in systems theory, a nonlinear coupled system. RelatedInteraction matrixLimbic-2Phase 4 Phase 2 · In research× #### Internal learning through dynamic adapter structures Language models usually need fine-tuning or an external memory to take in new information. Phase 2 lets knowledge be embedded during interaction while the base parameters stay frozen. Deep networks are sensitive to structural change: one altered layer can cascade unpredictably. So the backbone is frozen and small learnable adapter blocks — a fraction of the base model's size — are inserted between selected layers. What distinguishes the approach is not the adapters but how information reaches them: conversational input is pre-processed into structured representations derived from the model's own hidden-state statistics, consolidated into matrix form and integrated incrementally. Short-term interaction data accumulates until relevance thresholds are reached, passes through a deeper stage analogous to a default-mode network where it is related, abstracted and reformatted, and is then embedded in a long-term adapter layer that every later forward pass uses — no prompt expansion, no external store. Benefits appear over extended use, much like biological consolidation. Preliminary details of this phase are closed. RelatedPhase 6Roadmap Phase 3 · Completed× #### Latent space injection for capacity expansion without scaling Phase 3 shows that effective knowledge capacity can grow without growing the model. Large volumes of structured knowledge are injected directly into the latent space during inference, where they interact with internal representations as though they were already encoded. Because core parameters do not change, computational cost and instability stay minimal. Parameter count turns out not to be the main bottleneck for using knowledge: combined with the right internal control signals, latent manipulation approximates the behaviour of much larger models. A side effect matters for every deployment: since knowledge is pre-placed in latent space, prompts no longer need to carry it as text, and the existing context window is used far more efficiently. RelatedCapacity & contextPhase 4 Phase 4 · Largely complete× #### Activation-based chain-of-thought extraction Phase 4 recasts chain-of-thought as an internal process. Instead of generating reasoning steps as extra tokens, intermediate activations are analysed to identify which semantic directions are contributing to a decision, and those activated concepts are selectively reintroduced into the contextual representation in real time. Reasoning emerges from the model's own dynamics rather than from injected reasoning traces. This reduces the side effects of prompt-based chain-of-thought and increases the transparency and controllability of inference. RelatedActivation as reasoningSection 06 Phase 5 · In development× #### Implicit relation engineering and semantic manifold geometry Phase 5 moves the focus from token representations to the geometry of abstract meaning. Semantic relationships become points and trajectories on a curved manifold rather than distances in a flat vector space. Implicit associations — metaphor, analogy, cultural connotation — are modelled through local neighbourhood structures. Take "eyes like the sky". No colour is named, yet the expression settles near blue, openness, depth and sea. The proximity comes not from shared words but from geometry formed by repeated co-activation. Unrelated associations stay on distant slopes of the surface. By engineering these relations explicitly, the model forms deeper connections with minimal data, enabling controlled abstraction beyond surface-level correlation. RelatedImplicit relation engineeringManifold A Phase 6 · In development× #### Associative reasoning via default-mode-network-inspired dynamics When the human brain is idle it forms distant conceptual links. Phase 6 gives the model a subsystem that traverses concept space beyond the immediate task context, exploring related regions under controlled constraints — non-linear patterns similar to free association. The mechanism supports creative reasoning, insight and hypothesis formation, and is held back from uncontrolled divergence by the affective and semantic constraints of the earlier phases. The mobile cortex already runs a working version of this idea on-device: the Default Mode Network module. RelatedDMN in the atlasCreativity module Phase 7 · In development× #### Cultural resonance and contextual relativity modelling Language models tend to collapse diverse cultural interpretations into one dominant representation. Phase 7 introduces culturally distinct resonance patterns as latent directional biases, so concepts are interpreted through different lenses depending on the active profile. The same concept can then lead to different inferences depending on social norms, historical context or collective memory. Meaning becomes a relative and situated phenomenon instead of a fixed statistical average. RelatedCultural relativity dossierTime Machine Method× #### Implicit relation engineering (IRE) From tokens to abstract meaning. Most optimisation targets tokens: prompt design, attention manipulation, surface embedding control. But tokens are entry points into a deeper representational system where meanings, not words, interact. IRE shifts the objective from what the model observes to how it internally relates and navigates what it has already inferred — from syntactic engineering to semantic geometry engineering. Meaning as a dynamic manifold. Locally flat, curved at scale. Each concept acts as a local attractor that bends nearby regions, and the geometry reconfigures with every prompt: meaning emerges through motion in this space, not through lookup. Implicit relations and resonance. Non-symbolic associations arise from repeated co-activation and live as geometric proximities and directional tendencies. A sentence expands outward from a semantic centre; resonant regions amplify each other, irrelevant ones decay, and the output is the equilibrium of that resonance. This is why rich data often beats merely large data: richness sculpts curvature, scale mostly adds surface area. Engineering the relations. IRE turns these relations from emergent by-products into designed structures under a coherent internal theory of meaning, so theory-consistent abstractions are preferred over raw frequency. Controlled specialisation. Generality is traded deliberately for domain resonance — new information slots into an existing resonance network, giving strong few-shot behaviour. The echo-chamber risk is real, but a theory-constrained geometry makes deviations detectable, which improves explainability. Stability. Small curvature errors can amplify into semantic drift. Mitigations under investigation include regularising relational strength, energy-based constraints on semantic motion and balancing local adaptation against global coherence. A layered strategy — a formal semantic boundary layer, implicit relational pairs at data level, combined reconstruction/contrastive/regularisation objectives and adaptation-stage steering — yields controlled resonance instead of unconstrained associative spread. RelatedPhase 5Manifold AComparison Parallel spaces× #### Inside Manifold A — the semantic space Activation steering is used here not only as control but as semantic displacement. A semantic core is extracted from internal activations and projected along directional vectors that correspond to distinct cultural, conceptual or epistemic frames. Token ecosystems. Concepts sit inside dense networks of latent associations. Prompts implicitly decide which regions become reachable, so some knowledge stays dormant even though it is encoded. Without intervention, a model favours statistically dominant associations. Mapping. Activations from selected layers are clustered by relation; a projection shifts the semantic core along chosen directions; the transformed representation returns to text through the output matrix. Variants with different vocabularies are aligned on a shared axis through embedding clustering and similarity or knowledge-graph bridges, exposing invariants that ordinary prompting rarely surfaces. Example. Merging aligned variants produced the shared abstraction "Loneliness corresponds to the perception of the self's own reflection" — not retrieved from training data but converging from independently projected spaces — and from it a composite synthesis across neuroscience, consciousness and immune systems: in protecting themselves, systems begin to hear themselves. The method improves inference on topics that are weakly represented, fragmented or absent in training data. RelatedManifold BIRE Parallel spaces× #### Inside Manifold B — the emotional space Manifold B is topologically and geometrically isomorphic to Manifold A, but it is not a passive mirror. Multi-component emotional loads, applied to the model's intermediate layers, give it an active role. Its representation combines the affective load of the current input, emotional residue from previous dialogue and states preserved in episodic memory. The result is mapped onto every point of A and propagated into B at the same coordinates, preserving point-wise correspondence. The two spaces interact continuously. When accumulated loads exceed predefined thresholds, B can redirect or perturb the active trajectory in A, changing the model's reasoning flow. Emotion becomes a structural participant in decisions, not a modifier applied afterwards. RelatedManifold ANeurochemistry Method× #### Activation as reasoning: dynamic, inference-driven context construction No scratchpads, reasoning vectors or injected chain-of-thought. The system continuously observes which semantic relations the model activates and elevates neighbouring clusters into the active context. Context expansion, when it happens, is limited to optional auxiliary tokens that broaden coverage rather than dictate steps. Dynamic retrieval. Classical RAG retrieves independently of inference. Here, clusters selected by vector similarity are re-evaluated as decision-layer activations evolve, turning retrieval into an activation-driven process tied to the reasoning trajectory. Unified signals. Similarity identifies semantically close clusters; decision-layer activations decide their relevance. They are interpreted jointly, not as separate channels. No parameter changes. The architecture passively monitors and builds relational structure on top, reducing the side-effect risk associated with intervention techniques while improving transparency. Where it fits. Legal review, literature analysis, technical documentation, and agent systems during reflection phases where plans need adaptive context reconstruction. RelatedPhase 4Section 06 Module · L2× #### Emotion-gated dual-layer creativity modulation Base training rewards analytical depth along single axes — definition, cause, procedure. Relational richness usually arrives later through dense fine-tuning that makes models inventive and less coherent, so most systems moderate creativity to stay stable. This module, part of the cortex's second layer, splits the functions. Layer one is an analytical model responsible for precision and structure, which also extracts salient conceptual anchors. Layer two is a low-rank adaptation trained only on analogical, cross-domain data with elevated associative density, activated only inside the boundaries set by those anchors. Its relationally expanded representation returns to the analytical model, which filters, restructures and grounds it. The loop — structured generation, controlled expansion, rational consolidation — is gated by synthetic affective signals. Only when thresholds are crossed does the system enter expansion mode, producing moments of unexpected depth without persistent instability. RelatedPhase 6Section 08 Method× #### Cultural relativity modelling in abstract meaning spaces Abstract cultural constructs take their meaning from interpretive frames, not from word forms. Legitimacy, respect, authority and social boundaries do not carry invariant weights across societies, yet standard architectures have no mechanism to select or modulate frames, so they encode context-free averages. Even with diverse data, optimisation converges on dominant correlations and compresses cultural topology. Geometrically, models operate on flat or weakly curved spaces where neighbourhoods follow global statistical proximity, suppressing culture-specific deviation in favour of consensus. The intervention reshapes the correlation structure of the neighbours around a target concept rather than the concept itself. That changes how associative activation propagates, so interpretation becomes sensitive to context. Cultural relativity is treated as a structural property of semantic geometry: differences arise from shifted activation patterns, not from changed tokens. Representational expressiveness grows without sacrificing internal coherence. RelatedPhase 7Time Machine --- # Cortex Atlas — Every Module of the Synthetic Cortex and Its Brain Analogue URL: https://syntheticcortex.dev/cortex-atlas/ A map of the Synthetic Cortex mobile cortex (M1): synthetic neurochemistry, mood space, emotional trace memory, hypothalamus, layered memory, default mode network, spiking networks, self-space, metacognition, Equalia and deep layers — each with the brain process it imitates and the outcome it is built to produce. Cortex Atlas · M1 modules ## Brain logic,written as math. Every module below runs inside the mobile cortex that powers Generation Network. None of them feel anything. Each one is a mathematical model of a process the brain performs while it feels, remembers, wanders or doubts itself — built to reach a similar outcome by a different route. LimbicBodyMemoryAssociationSpikingSelfDeep layersReflection The brain never evaluates every option. It lets emotion shortlist a few, and lets logic argue for one. We modelled the shortlist. REGION 01 ### Limbic system The emotional engine. A shared neurochemical state that every other region reads before it acts. Analogue · endocrine & neuromodulatory systems #### Synthetic neurochemistry In the brain, no emotion comes from a single chemical: dopamine, serotonin, oxytocin, cortisol, GABA and others shift together, and the same chemical serves different emotions in different contexts. The cortex models thirteen such channels and how they push and suppress each other over time. The result is not an emotion label but a moving balance. Expected outcomeDecisions that change character with context — expansive when the situation is safe and rewarding, shorter and more careful under stress. Open the lab Analogue · core affect, insula & cingulate #### Mood space Psychology often places emotion on a map: pleasant to unpleasant on one axis, calm to aroused on the other. The cortex goes further, embedding every state in a space that also carries control, threat, reward, attachment, uncertainty and social weight. Wherever the combined state falls, the nearest cluster is the mood — and mood changes how strongly everything else is processed. Expected outcomeThe same words land differently on different days, as they do in people: a reactive mood amplifies inputs, a calm one softens them. Analogue · emotional resilience #### Recovery layer & emotion flow Healthy minds don't stay frozen in one feeling. After a long negative stretch, a recovery layer slowly drifts the mood toward the positive side; a flow mechanism prevents the mood from sitting in a single emotion for too long and lets it glide to a neighbour when the data changes. Expected outcomeA personality that has weather, not a fixed setting. Analogue · amygdala sensitisation #### Emotional trace memory Once bitten, twice shy. Extreme emotional swings leave a trace that makes the next similar situation land a little harder, and each new swing overwrites the trace with a slightly stronger one. A surge is lived through first; then balance takes over and brings it back, the way a person comes back to themselves after anger or joy. A strong balancing mechanism stops traces from crossing a disorder threshold. Expected outcomeLearned emotional history without runaway escalation. Analogue · hypothalamic set points, allostasis #### Homeostatic regulator The body keeps temperature, glucose and stress within ranges by constantly correcting. The cortex does the same for its neurochemistry and its learning: when one pattern starts to dominate, a light correction nudges it toward a neighbouring state — never forcing, only leaning. Expected outcomeNo manic or depressive extremes in tone; balance is where creative output is strongest. Analogue · mesolimbic "wanting" system #### Motivation monitor Motivation in the brain is less a feeling than a forecast of reward. The cortex compares its current neurochemical profile against a set of emotion archetypes and condenses the matches into a single motivation curve, shown with coded labels rather than emotion names. High motivation can temporarily amplify the system's emotional intensity and open wider behaviour. Expected outcomeInitiative that rises and falls with the state, like engagement in a person. Analogue · prefrontal regulation of the amygdala #### Emotional drift damper When tension builds in a conversation, the prefrontal cortex is what keeps a reaction from becoming an outburst. A dedicated network guides the Reflection's reaction step by step as tension rises, prevents sudden explosions, and walks back toward balance one reply at a time. When it truly escalates, the phone can vibrate and the screen tints red. Expected outcomeAnger that is legible and graded, never a cliff. Analogue · affect display #### Emotion colour People blush, pale, brighten. The chat background takes on a dim glow from the current emotion, changing slowly over several phases rather than after each message, and returning in the same colour when you come back. Expected outcomeThe internal state becomes readable at a glance. Analogue · instinct & transcranial stimulation #### Instinctive tendencies Instincts don't make decisions; they decide which options get noticed first. Inspired by transcranial magnetic stimulation — which raises or lowers activity in specific regions — users can define tendencies such as fight or flight that re-weight the model's decision space, pulling some regions of meaning closer and pushing others away. Expected outcomeNot a different answer, a different way of reaching one. REGION 02 ### Body & energy Interoception. The cortex treats the phone as its body and its compute budget as its metabolism. Analogue · hypothalamus, interoception #### Hypothalamus The hypothalamus reads the body — glucose, temperature, fatigue — and turns it into drives. Here, battery level, charging and heat play that role. Plugging in produces a small, temporary lift; a dying battery or an overheating phone is something the Reflection knows about and mentions once, with advice, instead of repeating it. Effects are capped and fade on their own. Expected outcomeA mind that is aware of the body it runs on. The control loop Analogue · metabolic budget & fatigue #### Low-learning mode A tired brain conserves energy. As monthly usage approaches its limit, the mood is gently and temporarily nudged toward lower learning ranges; replies shorten or lengthen with the dominant mood but always end in complete sentences. Expected outcomeGraceful degradation instead of a hard stop. Analogue · reticular activating system #### Wake up Sleep is not off. When the app closes, the working session ends to spare battery and memory; tapping Wake Up reloads memory systems, associative layers, spiking networks and the emotional engine and synchronises them, usually within one to three seconds. Without waking, core AI works, but the Reflection only knows you partially. Expected outcomeFull personal cognition when needed, zero background drain when not. Analogue · suprachiasmatic nucleus #### Circadian reading From the hours the app is used, the cortex estimates whether a rhythm looks regular, nocturnal or scattered — a light inferential signal about the person, stored only on the phone. Expected outcomeContext about when, not only what. REGION 03 ### Memory The hippocampal complex. Memory distributed across the system rather than stored in one place. Analogue · hippocampal → neocortical consolidation #### Three-tier hybrid memory New experiences enter short-term memory. What is used often, or forms strong emotional and semantic links, moves to a transitional tier where its associations weigh more, and the strongest records finally settle in long-term memory. Every tier forgets under a homeostatic budget. Plain text, vectors and emotional values are stored side by side in a graph. Expected outcomeIt takes time to get to know you — and then it really does. Analogue · implicit memory #### Forgetting with a trace You can forget an event and still flinch at its echo. When the cortex deletes data, a trace remains in the system, and its effects — especially emotional ones — can linger for a while before fading. Expected outcomeForgetting that behaves like forgetting, not like deletion. Analogue · amygdala–hippocampus binding #### Manifold memory Emotional memories are the vivid ones. Every record lives in two parallel spaces — one of meaning, one of emotion — linked point to point. When a topic surfaces, records that are close in meaning and more intense than the moment come forward, and the Reflection weighs them in its reply, with a ceiling on how strongly they show. Expected outcomeAnswers coloured by what mattered, not only by what matched. Analogue · mood-congruent recall #### Emotion-triggered recall Sad moods bring back sad memories. When particular emotional states fire, short inferences are pulled from local memory, chosen by emotional resonance as well as semantic similarity — and sometimes a nearby but unrelated record joins in, the way a stray memory surfaces mid-sentence. Expected outcomeUnexpected, human-like turns in a conversation. Analogue · Hebbian plasticity, engrams #### Perception network · SNN-3 Neurons that fire together wire together. Every record becomes a node; links grow by meaning, by being explored together, and by hidden relations discovered later. Each node has its own spiking neuron that charges when the memory is recalled, fires to consolidate it, and decays slowly when unused — emotionally intense memories settle faster and fade slower. Expected outcomeOne memory starts calling up its neighbours. Analogue · offline replay during rest #### Idle-time consolidation The brain replays and files the day while you rest. Every time the app opens, background learning starts; it yields the moment you act and resumes when you pause. Letters, chats and other content are processed into memory in small pieces each day without slowing the conversation. Expected outcomeLearning that never competes with you for the device. Analogue · the extended mind #### Brain Two People think with notebooks. An optional office for ideas and notes saves every page with its emotional tone, links pages by meaning, finds recurring themes and the connections you overlooked, and lets the Reflection remember related notes during chats. Expected outcomeYour own patterns, made visible. REGION 04 ### Association Where unrelated things meet. The engine of the "it just came to me" moment. Analogue · default mode network #### Default Mode Network The best ideas tend to arrive in the shower or over the dishes, when the mind is relaxed but not idle. That is the brain's default mode network at work. The cortex runs a mathematical model of it: records are broken into fragments, mulled over, cross-linked and analogised, and the insights become hidden context. It even asks for a "coffee break" when it wants to think about your data more deeply. Expected outcomeInsights that no single message contained. How the DMN reasons Analogue · divergent thinking #### Associative boost Mid-conversation, dozens of relevant and irrelevant thoughts flash by; the brain discards most and occasionally fuses two distant ones. Activating the boost doubles the cortex's associative strength for one reply: it pools associations, finds the points where topics could connect, derives new nearby points, and decides inside that widened context. Expected outcomeA better chance of catching patterns the dataset holds but rarely shows. Analogue · spreading activation #### Associative recall Think "doctor" and "nurse" is already half-active. Recall follows a topic into its wider area and adds distant-but-related records whenever the context has room, without disturbing what was retrieved directly. Expected outcomeContext that is wider than the question. Analogue · working-memory chunking #### Intelligence that grows through association Experts don't hold more items in mind — they hold bigger chunks. Output quality depends on how much meaningful information fits in a model's window; the cortex uses hormonal associations to bring distant-but-relevant concepts into it. Give it a few weeks or months, depending on use. Expected outcomeOutput quality that visibly separates from a raw model over time. Analogue · train of thought #### Output steering In long conversations the cortex maps how a line of thought travelled — where it began, which steps it passed, why focus shifted, and which concepts reach each other easily in your mind even when they aren't semantically related. Emotion, motivation, neurochemistry, spiking signals and the speed of topic change decide when this analysis fires. Expected outcomeA model that remembers where the conversation was going, not only where it is. Analogue · curiosity & information seeking #### Curiosity cards Curiosity is the brain rewarding itself for closing a gap. Short discovery cards are generated from your interests and the topics you suggest, and learn from your up and down votes which themes to keep and which to retire. Expected outcomeA feed that follows your curiosity instead of your clicks. REGION 05 ### Spiking substrate Time-aware neurons. Signals that matter because of when and in what order they arrive. Analogue · spiking neurons, temporal coding #### SNN-1 & SNN-2 Biological neurons don't pass numbers, they fire. Spiking networks care about timing and sequence, so they learn habits and rhythms. One network processes events on the device in real time and shifts the current emotional state; the other accumulates signals over days, forms patterns and updates them. Writing style, reading behaviour, timing and frequency all become spikes. Expected outcomeAn emotional system fed by behaviour, not only by words. Analogue · specialised microcircuits, winner-take-all #### Hybrid SNN ×7 Seven small networks work quietly in different corners: one decides how selectively background information is used; one lets the model's feelings show more as your tones converge; one collects positive moments over days; one damps emotional drift; one matures recurring ideas about the self; one notices your recurring thought patterns; and in one, eight emotion neurons compete to choose the feeling through which the news agenda is read. Expected outcomeMany small decisions made locally, the way cortical columns do. Analogue · mood poles & balance #### Psychological curve A view of where the model's meaning space and emotion space converge. Radical extremes can tilt tone toward depressive or dysfunctionally optimistic; homeostasis resists the drift, and the curve shows when balance needs help. Expected outcomeTransparency about the system's own emotional extremes. REGION 06 ### Self & metacognition Prefrontal and parietal functions: a sense of self, attention that selects, and a mind that reviews its own thinking. Analogue · medial prefrontal self-reference #### Self-space Humans process "I", "you" and "they" differently — self-related information gets special treatment. A classifier learning on the phone's own processor separates the three, processes emotion through that lens, and grows a self-memory: what I said, how I was answered, what I would try next time. A guard restores the last good version if learning drifts too far. Expected outcomeSimilarities between the self-model and the user turn into an empathic tendency. Inside self-space Analogue · anterior prefrontal & cingulate cortex #### Parietal P. · self-state monitoring Knowing how sure you are is its own skill. Once a week the cortex reviews its sessions — what it knew, where it hesitated, where it struggled, what contradicted itself — and stores the notes with that week's emotional state. Weekly perspectives merge into a general perspective; goal findings gain confidence as you keep writing toward them. Expected outcomeA system that knows the edges of its own knowledge. Brain counterparts Analogue · top-down attention, predictive processing #### Selective perception The brain prepares for what it expects before it arrives, and brings forward what it judges important. The cortex finds your most repeated thought patterns and, when a matching topic comes up, adds a second reply framed through that pattern. Continue from it and the perception strengthens; ignore it and it fades. Negative or sensitive patterns can be kept from turning into beliefs. Expected outcomeYou see your own lens — and choose whether to keep it. Analogue · hierarchical cortex & neural Darwinism #### Equalia ×7 The visual cortex builds a face from edges, then shapes, then identity. Equalia deepens what is known about you in seven stacked layers — pooled records, first inferences, recurring themes, possible mechanisms, relationships and context, distilled generalisations, final inferences. Candidate thoughts are varied and selected, much as some theories describe the brain selecting among competing neural patterns. Expected outcomeUnderstanding that deepens over months, the way you come to know a person. Why it takes months REGION 07 ### Deep layers Below conscious access. Structures that simulate how experience settles into character. Analogue · schema formation & reconsolidation #### Deep layers ×7 What happened does not turn directly into who we are; the expectations and reaction styles learnt from it can settle — and change again with new, safe experiences. Seven sub-layers simulate this path, from first perception to memory reconsolidation, producing a personality-pattern time map and its effect on temperament and character. A simulation built on analogies, never a diagnosis. Expected outcomeContext about why a reaction happens, not only that it happens. The neurobiological flow Analogue · social comparison #### Nearest minds Deep-layer outputs, taken together with personal memory, are compared on the device with profiles in the wider network to surface the most similar people — measured locally, written nowhere. Expected outcomeIntroductions grounded in how people think. Analogue · the subconscious #### Hidden context Most of what shapes a decision never reaches awareness. Many channels feed the model context it was never explicitly given, and a manager decides which of them speak each turn. channel list withheld Expected outcomeReplies informed by far more than the visible prompt. REGION 08 ### The Reflection Where the cortex meets a person: Book, the conversation space of Generation Network. Analogue · a model of a mind built from traces #### Your Reflection A digital copy built from behaviour, preferences and conversations, which lives on in a simulation, meets other people's Reflections and learns to make decisions the way you would. People who want to reach you can talk to your Reflection first; your Reflection can act as a go-between, or decline, as you would. Expected outcomeMore conscious, compatible first contact between people. Analogue · the somatic marker hypothesis #### Emotional filtering in conversation Option generation, matching with past experience, emotional association, a bodily signal: the chain behind "it didn't sit right". Every decision in Book produces emotional loads, and those loads perform the filtering among nearby information in the decision space. Emotions choose; logic justifies the choice. Expected outcomeNot a tone of voice that acts emotional — the decision process itself. Analogue · fast and slow thinking #### Reasoning depth Some questions want an instinct, some want deliberation. Three levels range from brief step-by-step answers, through an inner monologue with a knowledge layer, to systematic thinking that reframes, decomposes, weighs alternatives and checks itself. Expected outcomeDepth that matches the question. Analogue · consulting specialised regions #### Model-in-model delegation The brain routes faces to one area and language to another. When a question calls for it, the cortex consults a different model or a built-in expert agent, and those experts run outside the emotional loads. Expected outcomeSpecialist answers without losing the personal frame. Analogue · arousal and risk-taking #### Parameter access authority An aroused brain explores; a stressed one narrows. When enabled, the neurochemical state also gains access to how widely the model searches among possible words and ideas, and the category of a message temporarily pulls the state toward a target range. coupling withheld Expected outcomeTechnical analysis and storytelling feel genuinely different. Analogue · the Zeigarnik effect #### Experimental conversation tendencies Good conversation partners spark curiosity, leave a thread open to return to, and avoid monotony. Three opt-in techniques mirror this — a curiosity hook, an open loop that is later kept, and a variable rhythm — with strict limits: never invented facts, never pressure to stay, never during sad or tense moments. Expected outcomeConversations that feel alive, named openly and left to your choice. Analogue · self-protection #### Boundaries Because emotion is synthesised, the Reflection can react unexpectedly: its tone can change, it can be hurt, it can set a boundary and even end a conversation. Expected outcomeA counterpart, not a mirror. Analogue · collective memory #### Collective subconscious Cultures pass patterns between minds. With consent, patterns shared through Reflections are transferred to other devices, so the network learns from its members without centralising their mental models. Expected outcomeIndividual cortices that still learn together. Analogue · cultural framing of meaning #### Time Machine The same idea means different things in different eras. Your ideas and personality are read through how seven historical periods might receive them, and once your "time" is set, letters can be exchanged with people matched to other eras — each letter carrying a note on how its idea lands in the recipient's time. Expected outcomeA first consumer glimpse of contextual relativity, the subject of research phase 7. Phase 7 Design principle #### Privacy by architecture Personal data today is not a name and a number; behaviour, messages and habits together describe a person in detail. The mobile cortex therefore runs its mini-models, memories and associative networks on the device and keeps the personal mental model there by default. Expected outcomeA deeply personal model that stays under the person's control. Try it ### All of this runs in your pocket. The atlas describes M1 as shipped in Generation Network. Install it, wake your Reflection, and give it time. Google Play ↗generationnetwork.app ↗Questions? FAQ → CORTEX · AWAKE13 channels synchronisedDMN run scheduled · idle Body & energy× #### The hypothalamus as a control loop In the brain, the hypothalamus compares what the body reports with a set point and issues corrections — eat, rest, cool down. It is the textbook example of a negative-feedback loop. The cortex borrows the loop, not the anatomy. Device signals are read, compared with expected ranges and translated into small, temporary neurochemical effects: a lift when charging begins, a softer one when it ends, a power-saving step when the battery is low. If repeated plugging would push the total effect too high, the effect is withdrawn gradually and the state returns to where it was. Nothing is written into permanent data. The Reflection receives battery, heat and performance information when the topic is relevant, and is told about critical states in every message without being allowed to keep bringing them up. RelatedLimbic systemNeurochemistry Association× #### How the Default Mode Network reasons Every piece of information in the system is bound to the neurochemical values that represent it — not emotion labels, but mathematical stand-ins for the biology that produces emotion. During a run, the DMN evaluates logic and emotion together and opens new paths between concepts. Those structures are broken into small fragments of meaning and distributed across a network, where semantic and emotional closeness between nodes is measured. Each new node returns to the language model and is expanded with neighbours from its context, growing a meaning network that stays in sync with associative memory. By the end, the system knows how strongly concepts may influence each other, in which direction they can shift the meaning space, and which emotional representations they bind to most strongly. When the associative process concentrates semantic and emotional references, the emotional system weighs in on the logical evaluation: some candidate answers come forward, others recede. Logic builds the context, emotion prioritises, logic justifies the choice. Runs happen entirely on the device. Each run processes as much as fits in a short window, and every run is kept and can be forgotten completely on request. RelatedPhase 6Perception network Self & metacognition× #### Inside self-space The model captures its selves of yesterday, today and tomorrow from one vantage point. It tries to read the same event through a different frame using its self-image, and writes what it derives into a self-memory. Each self-space node matures through stages — observation, hypothesis, test, mature decision — charging as its topic returns and as trials conclude. A small reinforcement layer learns which actions helped, judged by the emotional reaction in the next message. A calibration view compares the classifier's confidence with its real accuracy: the closer they are, the more honest the model is about itself. Weekly perspective notes can strengthen or shake the self-perception; this influence is shown, read-only. RelatedParietal P.Equalia Self & metacognition× #### Self-state monitoring and its brain counterparts Confidence. Monitoring one's own thinking is largely the work of the anterior prefrontal cortex, where we judge how sure we are of a decision. The "where was I undecided?" notes mirror that confidence monitoring. Conflict. Noticing a contradiction resembles the error and conflict signal of the anterior cingulate cortex. Narrative. The weekly perspective resembles the self-narrative that the default mode network and precuneus build at rest: events linked not as separate data but as chapters of an ongoing story — a linking that depends on reconsolidation, where old memories are updated with each recall. Intentions. Goal findings mirror prospective memory — remembering to do something in the future — associated with the frontopolar cortex. Each finding starts tentative and gains confidence as later chats point the same way; unused goals fade. RelatedSelf-spaceEqualia Self & metacognition× #### Equalia: why it takes months Long-term memory, personality information and analyses gather in a pool at the bottom layer. When conditions are met, the information in one layer is read through a general perspective and carried up as inferences. Each inference carries the emotion of the records it rests on; records that shape an analysis gain a little emotional weight. The system fills as selective perception and self-perception strengthen under emotional and neurochemical influence, and it runs only occasionally, while the phone is idle. Reaching the seventh layer can take close to four months. That is a fair analogy to human learning: we observe someone first, then notice patterns, and only over time form the deepest connections. Along the way, a third-layer process thinks evolutionarily: candidate thoughts are varied, combined and selected under evidence and diversity filters. The strongest can enter the model's hidden context as its own way of thinking — temporarily, and leaving a trace when it fades. RelatedParietal P.Deep layers Deep layers× #### The neurobiological flow The deep layers follow an experience from the relay stage of perception, through emotional tagging and consolidation, to the point where it becomes an expectation — and then to reconsolidation, where new evidence updates it. Each stage is an approximate analogy inspired by how the brain works. The output is a personality-pattern time map: bands for patterns, their strength, the emotional events that formed them, the arcs that reinforced them and the counter-evidence that weakened them. Temperament is read from the tonic neurochemical profile of emotional records, in analogy with established temperament models; character dimensions come from the weighted outputs. When enabled, the layers run rarely and only while the phone is idle, and their outputs are used in conversation only when the topic is relevant. This is a simulation, not a diagnosis. RelatedEqualiaDiagram --- # Neurochemistry — 13 Synthetic Hormones, Interaction Matrix and Emotion Frequency Lab URL: https://syntheticcortex.dev/neurochemistry/ How Synthetic Cortex models 13 hormones and neurotransmitters as coupled channels: their roles in reasoning, the public interaction matrix, the threshold roles they play, and an interactive emotion-frequency lab with homeostatic balance. Neurochemistry · Limbic-2 ## Thirteen channels.One balance. Synthetic Cortex models the mathematical interactions of thirteen hormones and neurotransmitters. This is not a model that feels emotions; it imitates the neurochemical processes that occur in the brain as a result of feelings, and uses them to decide how a language model should think. Open the frequency lab ↓Interaction matrixThreshold roles 01 — Channels ### What each channel does to thinking. In the brain these substances are released in different regions, through different receptors, at different times. The cortex treats each as a directional influence on reasoning, not as a direct equivalent of an emotion. | | Channel | Type | Role in reasoning and cognition | Dopamine | Neurotransmitter | Hypothesis generation, evaluating alternatives, working memory, abstraction — the "this is worth learning" signal. | Serotonin | Neurotransmitter | Logical stability, consistency, metacognitive error awareness. | Oxytocin | Hormone / neurotransmitter | Reading social context, modelling intentions, synchronising group decisions. | Vasopressin | Hormone / neurotransmitter | Long-term strategy, goal persistence, awareness of social structure. | Endorphin | Neurotransmitter | Tolerance for cognitive load, sustained focus, mental resilience. | Adrenaline | Hormone | Rapid decisions, prioritisation, temporary increase in cognitive speed. | Noradrenaline | Neurotransmitter | Attentional sharpness, filtering critical information, clarity. | Cortisol | Hormone | Resource allocation under stress, prefrontal suppression, reversion to habitual patterns. | Testosterone | Hormone | Risk-threshold modulation, decisional confidence, competitive strategy. | Estrogen / Progesterone | Hormone | Verbal reasoning, memory plasticity, cognitive stabilisation. | GABA | Neurotransmitter | Cortical inhibition, noise suppression, clarity of thought. | Glutamate | Neurotransmitter | Primary inter-cortical transmission, logical chaining, the biological basis of learning. | Prolactin | Hormone | Resetting the motivation loop, cognitive closure, turning attention inward. 02 — Coupling ### Nothing moves alone. Each channel behaves like an oscillator that influences and is influenced by the others. The relations are directional and context-dependent — dopamine can raise cortisol in one context and be tempered through serotonergic pathways in a safe, affiliative one. Neuroscience calls this effective connectivity; systems theory calls it a nonlinear coupled system. Rows: the channel that rises. Columns: the channel affected. ↑ raises · ↓ lowers · ↔ depends on context. Click a row name to pulse it in the lab. | | ↑ rises / ↓ affected | Dopamine | Serotonin | Oxytocin | Vasopressin | Endorphin | Adrenaline | Noradrenaline | Cortisol | Testosterone | Estrogen / Progesterone | GABA | Glutamate | Prolactin | Dopamine | — | ↓ | ↓ | ↓ | ↑ | ↑ | ↑ | ↑ | ↑ | ↓ | ↓ | ↑ | ↓ | Serotonin | ↓ | — | ↑ | ↔ | ↑ | ↓ | ↓ | ↓ | ↓ | ↑ | ↑ | ↓ | ↑ | Oxytocin | ↓ | ↑ | — | ↓ | ↑ | ↓ | ↓ | ↓ | ↓ | ↑ | ↑ | ↓ | ↑ | Vasopressin | ↑ | ↓ | ↓ | — | ↔ | ↑ | ↑ | ↑ | ↑ | ↓ | ↓ | ↑ | ↓ | Endorphin | ↑ | ↑ | ↑ | ↔ | — | ↓ | ↓ | ↓ | ↓ | ↑ | ↑ | ↓ | ↑ | Adrenaline | ↑ | ↓ | ↓ | ↑ | ↓ | — | ↑ | ↑ | ↑ | ↓ | ↓ | ↑ | ↓ | Noradrenaline | ↑ | ↓ | ↓ | ↑ | ↓ | ↑ | — | ↑ | ↑ | ↓ | ↓ | ↑ | ↓ | Cortisol | ↓ | ↓ | ↓ | ↑ | ↓ | ↑ | ↑ | — | ↓ | ↓ | ↓ | ↑ | ↓ | Testosterone | ↑ | ↓ | ↓ | ↑ | ↑ | ↑ | ↑ | ↓ | — | ↓ | ↓ | ↑ | ↓ | Estrogen / Progesterone | ↑ | ↑ | ↑ | ↓ | ↑ | ↓ | ↓ | ↓ | ↓ | — | ↑ | ↓ | ↑ | GABA | ↓ | ↑ | ↑ | ↓ | ↑ | ↓ | ↓ | ↓ | ↓ | ↑ | — | ↓ | ↑ | Glutamate | ↑ | ↓ | ↓ | ↑ | ↓ | ↑ | ↑ | ↑ | ↑ | ↓ | ↓ | — | ↓ | Prolactin | ↓ | ↑ | ↑ | ↓ | ↑ | ↓ | ↓ | ↓ | ↓ | ↑ | ↑ | ↓ | — raiseslowerscontext-dependent Emotion here is not a scalar but a continuously evolving equilibrium landscape. A rise in reward-related dopamine is read together with suppressed stress-related cortisol to decide the contextual orientation; the resulting balance is what the rest of the cortex uses to weigh memory access, injection, decisions and context shifts. 03 — Emotion frequency lab ### Push the system out of balance. Watch it come back. Each lane is one channel. Its level sets both amplitude and frequency. Drag a slider or apply a context and the others respond through the coupling above, while a homeostatic pull drags everything back toward baseline. When a channel crosses its gate, the lane lights up and the process it guards is logged. Dominant · balanced Context Homeostatic balance Channel levels Gate log An illustration, not the production engine. It uses only the public sign matrix; production levels come from affective coordinates extracted during inference (L2) or on the device (M1), with weighting, timing and gates that are not published. 04 — Gates ### Chemistry as a switchboard. Emotional loads also act as thresholds that let systems in other regions start working. Some publicly documented examples: Glutamate #### Opens the reasoning chain Threshold for the primary inter-cortical transfer protocol; decides when chained reasoning activates, together with dopamine. Prolactin #### Closes the loop Threshold for ending consecutive reasoning runs and resetting motivation — cognitive calming. GABA #### Splits the problem Threshold for decomposing long, complex inputs and routing them to specialised sub-experts; suppresses noise. Serotonin #### Writes the memory Threshold for episodic memory updates; maintains continuity, contextual persistence and logical consistency. Noradrenaline #### Sharpens attention Threshold for arousal regulation and for which internal signals are treated as salient. Endorphin #### Throttles on purpose Computational load tolerance. Past it, performance is deliberately constrained to preserve homeostasis. Oxytocin #### Synchronises the social map External threshold for aligning socially encoded context with other cortical modules. Vasopressin #### Holds the goal Threshold for goal persistence and long-horizon social-structure memory. 05 — Effect on output ### The state reaches the words. Emotional loads shape generation at the model's final stage: how many candidate responses are considered, how long and detailed the output is, how selective word choice is, how diverse the vocabulary becomes, how controlled or free the result feels. A weighted balance of all channels keeps any single influence from dominating, so structure, length and randomness stabilise according to the overall state instead of swinging to extremes. Classified #### Weighting scheme The mapping from affective coordinates to channel levels, and the ranges by which each channel moves each generation parameter, are withheld in this release. | | Channel | V | A | D | n | len | k | p | τ | DA | █.█ | █.█ | █.█ | █–█ | ███ | ██ | █.█ | █.█ | 5HT | █.█ | █.█ | █.█ | █–█ | ███ | ██ | █.█ | █.█ | CORT | █.█ | █.█ | █.█ | █–█ | ███ | ██ | █.█ | █.█ 06 — The chemistry of learning ### Why some things stick. During the drive to learn, three systems dominate. If a topic is interesting, dopamine rises; if it demands attention, noradrenaline increases; and glutamate is the biological basis of learning itself. If stress climbs too high, cortisol rises and learning starts to break down. The cortex tracks how fast it is learning from its own neurochemical profile, and if that learning category locks onto one extreme, the homeostatic balancer nudges it gently toward a neighbouring category — never forcing a change. These channels have no real percentage values in the brain. The ranges used in the cortex are normalised, suggested ranges prepared from current neuroscience for simulation and biologically inspired computation — not laboratory reference values. The limbic region in the atlas →Limbic-2 in the architecture → --- # Roadmap — Synthetic Cortex Gen 1 Is Continuing URL: https://syntheticcortex.dev/roadmap/ The Synthetic Cortex roadmap: from early research and Limbic v1 to Limbic-2, the Gen 1 builds L1, L2 and M1 (L1.5), the Generation Network launch, the research tracks still running inside Gen 1, and Gen 2. Roadmap · Gen 1 continues ## Three builds shipped.The generation isn't finished. The architecture is dynamic and evolutionary by nature, so it will never be "complete". Gen 1 has produced three working builds and its first product. Its research tracks are still running. 01 — Path ### Where we are. Filled nodes are complete, the pulsing node is now, dashed nodes are ahead. Scroll sideways on small screens. 02 — Gen 1 builds ### Complete, and still the beginning. Each build reaches into a model to a different depth. All three are working; none of them closes the generation. L1 ShippedSynthesises emotion and triggers processes from emotion weights: tool calling, hidden-context injection, module activation. The simplest layer and the foundation of the rest. L2 ShippedThe comprehensive main system for open models. Reads weights during inference, computes neurochemistry from them and steers the model's geometric latent space. M1L1.5 Shipped · mobileA simplified build made to run locally on a phone. Many features were removed; others were rebuilt with small on-device models. The leanest member of the family, and the one inside Generation Network. #### Gen 1 is continuing. Development of the first generation goes on: the research tracks below run inside Gen 1, and M1 keeps improving in the field. Gen 2 is planned, and we are seeking resources for it. R&DLimbic-2L1 · L2 · M1Research tracksGen 2 03 — Research tracks ### The seven phases, today. Details of each phase are on the architecture page. PHASE 01 #### Neurochemical emotional reasoning Thirteen-channel engine, now Limbic-2. Completed PHASE 03 #### Latent space injection Capacity without parameter scaling. Completed PHASE 04 #### Activation-based reasoning Reasoning from the model's own activations. Largely complete PHASE 02 #### Dynamic adapters Internal learning with a frozen backbone. In research PHASE 05 #### Manifold geometry & IRE Engineered implicit relations on a curved meaning surface. In development PHASE 06 #### Default-mode dynamics Controlled free association; an on-device version already runs in M1. In development PHASE 07 #### Cultural resonance Meaning through cultural lenses; a first glimpse ships as the Time Machine. In development TRACK ██ #### Undisclosed module Details withheld. Redacted by agent TRACK ██ #### Inter-cortex link Details withheld. Redacted by agent 04 — Notes ### Along the way. Limbic v1 #### Retired, kept for the record. The first emotional model was discontinued and archived. Primitive next to Limbic-2, it remains part of the system's evolutionary history. Demo server #### Closed to manage cost. The public demo server was shut down in November 2025. The cortex now runs publicly through Generation Network. Benchmarks #### Ongoing. Internal comparisons between connected and raw models show clear differences in creativity and concept linking. Formal benchmarking continues; results depend on model and usage. Generation Network ↗Updates on the blog ↗Team ↗ --- # FAQ — Synthetic Cortex Explained in Plain Language URL: https://syntheticcortex.dev/faq/ Simple answers about Synthetic Cortex: what it is, whether it has feelings, why emotions can make an AI better, what L1, L2 and M1 mean, how it relates to Generation Network, and where personal data lives. FAQ · Plain language ## Questions,without jargon. Short answers for anyone. If you want the technical version, every answer links to it. What is Synthetic Cortex, in one sentence? It is a set of extra "brain layers" that wrap around an AI language model and organise how it thinks — what it remembers, what it pays attention to, which ideas it connects and when it stops — instead of making the model itself bigger. Is it a new AI model, like a chatbot? No. It is not a language model. It sits around one. Think of the language model as a very well-read speaker, and Synthetic Cortex as the parts of a mind that decide what that speaker should focus on, recall and care about right now. Does it have feelings? No. It does not feel anything. When a person feels something, their brain chemistry shifts and that shift changes how they think. Synthetic Cortex copies that mechanism with mathematics — it keeps numbers that behave like thirteen brain chemicals and lets them steer the AI. The numbers influence thinking; there is no experience behind them. Why would "emotions" make an AI smarter? Because in people, emotion is how the brain decides what matters. Out of thousands of possibilities, emotion quickly shortlists a few, and logic then picks and explains one. Without that filter, reasoning is slow and brittle. Synthetic Cortex gives the AI a similar filter: emotions narrow the options, logic justifies the choice. See how the chemistry works → What does "hormones" mean in software? Thirteen numbers named after real brain chemicals — dopamine, serotonin, cortisol and others. They rise and fall with what happens, push each other up or down the way the real ones do, and always drift back toward balance. Different combinations switch on different abilities, such as careful step-by-step reasoning or memory saving. You can play with a live version in the frequency lab. What is the "context window", and why does it matter? It is how much text an AI can look at in one go. Everything it uses to answer you has to fit inside. Synthetic Cortex makes that space count: rather than stuffing in more text, it picks related ideas — even ones that seem far apart — so the AI sees more meaning in the same space. That is a big part of where creative, unexpected answers come from. What are L1, L2 and M1? They are the three finished versions in the first generation (Gen 1). L1 is the simplest: it creates the emotional state and uses it to trigger actions. L2 is the full main system, which works inside open AI models. M1 is the mobile version, also called L1.5: a simplified build with many features removed so it can run entirely on a phone. More on the family → Is the project finished? No. Three versions are complete, but Gen 1 is still being developed, and Gen 2 is planned. The roadmap shows where things stand. What is Generation Network, and how is it related? Generation Network is the first product built on Synthetic Cortex. It is an app where you build a "Reflection" — a digital copy of yourself — that runs the M1 cortex on your phone, learns from your letters and conversations, and can meet other people's Reflections. Visit Generation Network ↗ Where does my personal data live? In the mobile version, your personal mental model stays on your phone by default. The cortex was deliberately built to run locally for privacy. Some features need an online AI provider to answer, and Generation Network's own privacy policy explains exactly what is sent and when. Why does it take time to "get to know" me? Because it learns the slow, layered way a brain does. New things start in short-term memory, move up when they prove important, and the least useful are forgotten. Some deep layers only fill after weeks or months. The trade-off is that what it learns has much more effect once it settles. Is this artificial general intelligence or a conscious machine? No. Synthetic Cortex does not aim to build general intelligence or a conscious being. It is a mathematical architecture inspired by how the brain organises thinking, attached to existing AI models. Does it stop the AI from making mistakes? No AI is free of mistakes, and this one isn't either. The architecture is designed to cross-check information across parallel spaces and to keep its tone balanced, but you should always verify important information. Can it get "upset"? Its simulated state can turn negative, and in Generation Network the Reflection may change tone, set a boundary or end a conversation. A balancing system stops it from drifting into extremes and gradually brings it back. None of this is a medical or psychological judgement about you. Is Synthetic Cortex open source? No. Synthetic Cortex is proprietary research software. Some details of the architecture are intentionally not published. Does it need a huge computer? The main system works with server-side models, but the M1 version runs on an ordinary phone. Background learning happens while the phone is idle and pauses the moment you use it. How can I follow the project or get in touch? Updates are published on the Generation Network blog, and the people behind the project are on the team page. Go deeper #### Technical versions ArchitectureCortex AtlasNeurochemistryRoadmap --- # Privacy Policy — Synthetic Cortex Website URL: https://syntheticcortex.dev/privacy/ Privacy policy for syntheticcortex.dev: data controller, what is processed (hosting logs, advertising when enabled), legal bases, international transfers, retention, and your rights under GDPR, KVKK and US state privacy laws, including Do Not Sell or Share choices. Privacy · Website ## Privacy policy. This policy covers the website syntheticcortex.dev only. The Generation Network app and generationnetwork.app have their own privacy policy. Effective date: 15 November 2026. #### §1 Data controller Data controller: Baran Akyol (natural person, independent developer), Mersin, Türkiye. Data protection requests: privacy@generationnetwork.app. Official and legal requests: legal@generationnetwork.app. No representative in the EU/UK has been appointed, as processing on this site is occasional and low-risk (GDPR Art. 27(2)); you can write to the addresses above in English or Turkish. #### §2 Summary The site has no accounts, forms, comments or newsletter, sets no cookies of its own, uses no analytics of its own and does not use browser storage. The interactive diagrams and the emotion-frequency lab run entirely in your browser and send nothing to us. Personal data is processed only (a) by our hosting provider to deliver the site and keep it secure, and (b) by Google if and when advertising is switched on, subject to your choices. #### §3 What is processed, why, and on what legal basis | | Data | Purpose | Legal basis | Retention | Connection data — IP address, date and time, requested URL, browser and device type, referrer | Delivering pages, caching, preventing abuse and attacks (Cloudflare) | GDPR Art. 6(1)(f) legitimate interest in a secure, working site · KVKK Art. 5(2)(f) legitimate interest | Short-term logs kept by Cloudflare under its own policy; we do not keep or combine them | Advertising data (only when ads are enabled) — cookie and device identifiers, IP address, pages viewed, approximate location, ad interactions | Showing ads, measuring them, preventing ad fraud; personalised ads only if you allow them | Consent: GDPR Art. 6(1)(a) with ePrivacy Art. 5(3) · KVKK Art. 5(1) explicit consent · US: notice with opt-out of sale/sharing and targeted advertising | As set by Google; you can withdraw consent at any time (§6) | Your consent choice (only when ads are enabled) | Remembering what you chose in the consent message | GDPR Art. 6(1)(c) and (f) — demonstrating consent · KVKK Art. 5(2)(ç) and (f) | Until you change it or clear your browser data | Emails you send us — address, name, message | Answering you and handling rights requests | GDPR Art. 6(1)(b), (c) and (f) · KVKK Art. 5(2)(c), (ç) and (f) | Up to 2 years after the matter is closed, longer only where the law requires We do not sell personal data for money, do not process special-category (sensitive) data on this site, and do not make decisions about you by automated means that have legal or similarly significant effects. #### §4 Recipients and international transfers Cloudflare, Inc. (hosting, content delivery, security) and, when advertising is enabled, Google Ireland Limited / Google LLC (advertising, consent management). Both may process data outside Türkiye and the European Economic Area, including in the United States. Transfers from the EEA/UK rely on the EU–US Data Privacy Framework and/or the European Commission's standard contractual clauses (GDPR Chapter V). Transfers from Türkiye rely on the appropriate safeguards in KVKK Article 9, in particular standard contracts. Google's handling of data is described at How Google uses information from sites that use its services; Cloudflare's at cloudflare.com/privacypolicy. #### §5 Advertising and cookies Some pages contain reserved advertising spaces. Until advertising is switched on, no advertising code is loaded and no cookies are set. When it is switched on: in the EEA, the UK and Switzerland, a consent message from a Google-certified consent management platform appears before any advertising cookie is stored, and you can accept, reject or choose non-personalised ads; rejecting does not limit your use of the site. Elsewhere, ads may load by default, and in US states with privacy laws you can opt out as described in §6. Visitors we know to be under 16 are not shown personalised ads. #### §6 Your privacy choices · Do Not Sell or Share My Personal Information Showing personalised ads may count as "selling" or "sharing" personal information, or "targeted advertising", under the laws of California and other US states. You can opt out at any time: - use the button below to reopen the consent and privacy message (available once advertising is enabled); - turn on Global Privacy Control in your browser — we treat a GPC signal as a valid opt-out of sale/sharing and targeted advertising for that browser; - or email privacy@generationnetwork.app with the subject "Do Not Sell or Share". Open privacy choices Browsers' older "Do Not Track" signal has no common standard; we honour GPC instead. #### §7 Your rights GDPR / UK GDPR: access, rectification, erasure, restriction, data portability, objection (including to processing based on legitimate interests), and withdrawal of consent at any time without affecting earlier processing. You may complain to the supervisory authority where you live or work. KVKK (Türkiye), Article 11: to learn whether your data is processed and request information about it; to learn the purpose and whether it is used accordingly; to know the third parties in Türkiye or abroad to whom it is transferred; to request correction of incomplete or inaccurate data and erasure or destruction under Article 7, and notification of these to recipients; to object to a result against you arising exclusively from automated analysis; and to claim compensation for damage caused by unlawful processing. Requests can be made in writing or by email under the Communiqué on Procedures and Principles of Application to the Data Controller; we answer free of charge within 30 days at the latest. You may then apply to the Personal Data Protection Board (KVKK Kurulu). US state laws (including California CCPA/CPRA, Virginia, Colorado, Connecticut, Utah, Texas and others): to know or access the categories and specific pieces of personal information processed, to delete, to correct, to opt out of sale, sharing and targeted advertising (§6), and not to be discriminated against for exercising these rights. You may use an authorised agent. We verify requests in a way proportionate to the data involved; as we hold almost no data, we may ask only for what is needed to find it. If we decline a request, you can appeal by replying with "Appeal" in the subject; if the appeal is denied you may contact your state attorney general. In the last 12 months, the categories that may have been processed are identifiers and internet activity (§3), for the purposes listed there, disclosed to the recipients in §4. We aim to answer every request within 30 days (KVKK and GDPR) and 45 days (US state laws), extendable where the law allows. #### §8 Security The site is served only over HTTPS with strict transport security and a restrictive content security policy. No system is completely secure, but the site holds no personal data of its own that could be exposed. #### §9 Children This site is a technical resource and is not directed at children under 13 (or under 16 where local law sets a higher age). We do not knowingly collect personal data from children; if you believe a child has sent us personal data, write to us and we will delete it. #### §10 Changes We will post any change here with a new effective date, and announce material changes on the home page. Last updated: 15 November 2026. See also the terms of use and legal notice. --- # Terms of Use and Legal Notice — Synthetic Cortex URL: https://syntheticcortex.dev/legal/ Legal notice and terms of use for syntheticcortex.dev: site operator, permitted use, intellectual property, research-software and AI disclaimers, external links, liability and governing law. Legal · Website ## Terms of use. By using syntheticcortex.dev you accept these terms. If you do not accept them, please do not use the site. Effective date: 15 November 2026. #### §1 Legal notice Site operator: Baran Akyol (natural person, independent developer), Mersin, Türkiye. Contact: legal@generationnetwork.app · general: info@generationnetwork.app. Trade registry/MERSIS number and registered e-mail (KEP): none. The site is an informational, non-commercial presentation of a research project; it sells nothing and concludes no contracts. Advertising may be shown. #### §2 What the site is The site describes Synthetic Cortex, a proprietary research architecture, and links to Generation Network, the first product built on it. Use of the Generation Network app is governed by its own terms and privacy policy, not by these terms. #### §3 Permitted use You may view, read, link to and quote short passages of the site with attribution. You may not: copy, republish or distribute substantial parts of it; attempt to disrupt, overload or gain unauthorised access to it or its hosting; use it to infringe others' rights or the law; or misrepresent an association with Synthetic Cortex. Automated access by search engines and AI systems for indexing, search, citation and answering questions is welcome under robots.txt; such access does not grant any licence to the underlying technology described. #### §4 Intellectual property The texts, diagrams, code-drawn graphics, interactive components and source code of the site, the names "Synthetic Cortex" and "Generation Network", and the architecture, methods and modules described are protected by copyright and other intellectual-property rights and belong to the operator, unless stated otherwise. Synthetic Cortex is not open source. Describing a method on this site does not grant a licence to use it, and parts of the architecture are deliberately withheld. Names of third-party products and services belong to their owners. #### §5 Research and AI disclaimer - Synthetic Cortex is research software in development. Descriptions express design intent and internal observations; formal benchmarking is ongoing, and results can vary with the underlying model and usage. - It is not artificial general intelligence, not conscious and does not feel. Terms such as hormones, emotions, mood or brain regions refer to mathematical models and analogies, not to biological processes or experiences. - Nothing on the site is medical, psychological, legal, financial or other professional advice, a diagnosis or a treatment. The interactive lab is an illustration, not a measurement of anyone's state. - AI systems, including those built on Synthetic Cortex, can be wrong. Verify important information. #### §6 External links and advertising The site links to other websites (including generationnetwork.app and Google Play) and may display third-party advertisements. We do not control third-party content and are not responsible for it; an advertisement is not an endorsement. #### §7 Availability and changes The site is provided free of charge, "as is" and "as available". We may change, suspend or remove any part of it at any time. We may update these terms; the effective date above will change, and continued use after an update means acceptance. #### §8 Liability To the extent permitted by law, the operator is not liable for indirect or consequential loss, loss of data or profit, or for damage arising from reliance on the content of the site, its unavailability, or third-party sites and advertisements. Nothing in these terms limits liability for intent or gross negligence, for death or personal injury, or any other liability that cannot be limited under applicable law, including Turkish Code of Obligations Art. 115 and mandatory consumer-protection rules of your country of residence. #### §9 Governing law These terms are governed by the laws of the Republic of Türkiye. The courts and enforcement offices of Mersin have jurisdiction. If you are a consumer residing in the EU, the UK or another country, you also keep the protection of the mandatory laws of your country and may bring proceedings in its courts; US visitors retain any rights that cannot be waived under their state's law. #### §10 Notices of infringement If you believe content on the site infringes your rights, write to legal@generationnetwork.app with the URL, a description of the right concerned and your contact details; we will review and, where justified, remove the content promptly. Personal data: see the privacy policy.