CBA is an AI agent framework built on the foundations of over 50 neuroscience and cognitive science publications. Rather than treating the brain as a loose metaphor, every region, memory layer, and neuromodulator pathway in this system traces back to peer-reviewed research — from Baars' Global Workspace Theory to McClelland's Complementary Learning Systems, from LeDoux's amygdala fast-path to Hickok & Poeppel's dual-stream language model.
The architectural direction was initially inspired by OpenClaw. We studied its modular design philosophy and adapted it into a neuroscience-grounded cognitive pipeline, optimizing each component to mirror how the human brain actually processes information — from sensory gating through emotional appraisal to speech production.
What CBA has evolved into. The neuroscience-grounded architecture is the foundation, not the final product. CBA now includes a lossless knowledge curator for coding agents — a place where you drop multimodal business-logic information (text, images, PDFs, audio) and get back structured, workspace-partitioned, contradiction-aware context that downstream coding agents (Claude Code, Cursor, etc.) can consume via MCP. The biological architecture gives us principled answers to hard questions: what to forget, when to ask clarifying questions, how to detect contradictions, how to separate similar events. See the Knowledge Layer section below for the completed Phase 0-8 implementation.
This project is far from complete. There are rough edges, unexplored ideas, and plenty of room for improvement. We are releasing CBA as open source with the hope that it can grow through community collaboration — researchers, engineers, and curious minds contributing perspectives we haven't considered, catching mistakes we've overlooked, and pushing the framework in directions we haven't imagined. If even a small part of this work sparks a useful conversation or inspires a new approach, it will have been worthwhile.
Contributions, feedback, and discussion are always welcome.
- Python >= 3.11
- Node.js >= 18 (for dashboard)
- API key from any supported provider
git clone https://github.com/hyungwoo822/CBA.git
cd CBA
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e ".[dev]"
# Dashboard
cd dashboard && npm install && cd ..cp .env.example .envSet at least one API key in .env:
| Provider | Env Variable | Model Example |
|---|---|---|
| OpenAI | OPENAI_API_KEY |
openai/gpt-4o-mini (default) |
| Anthropic Claude | ANTHROPIC_API_KEY |
anthropic/claude-sonnet-4-20250514 |
| Google Gemini | GEMINI_API_KEY |
gemini/gemini-2.0-flash |
| xAI Grok | XAI_API_KEY |
xai/grok-2 |
Override the default model:
BRAIN_AGENT_MODEL="anthropic/claude-sonnet-4-20250514"from brain_agent import BrainAgent
async with BrainAgent() as agent:
result = await agent.process("Explain how memory consolidation works")
print(result.response)brain-agent run # Interactive agent
brain-agent dashboard # Start dashboard (port 3000)
brain-agent memory stats # Memory statistics%%{init: {'theme':'base', 'themeVariables': {
'primaryColor': '#F5F1E4',
'primaryTextColor': '#5C5847',
'primaryBorderColor': '#A8A078',
'lineColor': '#B8A5C9',
'secondaryColor': '#E8DFF0',
'tertiaryColor': '#F0E8D8',
'fontFamily': 'Georgia, serif'
}}}%%
flowchart TB
UserInput([User Input]):::inputStyle
subgraph Phase1[" Phase 1: SENSORY INPUT "]
direction LR
Thalamus[Thalamus<br/>relay]:::sensoryStyle
VisCtx[Vis Crtx<br/>Aud L+R]:::sensoryStyle
SensoryBuffer[Sensory<br/>Buffer]:::sensoryStyle
Thalamus --> VisCtx --> SensoryBuffer
end
subgraph Phase23[" Phase 2+3: DUAL STREAMS + INTEGRATION <br/><i>Hickok & Poeppel 2007</i> "]
direction TB
subgraph Ventral["Ventral Stream semantic"]
Wernicke[Wernicke LLM<br/>comprehension]:::ventralStyle
Amygdala[Amygdala LLM<br/>R=fast, L=ctx]:::ventralStyle
end
pSTS{{pSTS merge}}:::mergeStyle
subgraph Dorsal["Dorsal Stream motor"]
Spt[Spt<br/>auditory-motor]:::dorsalStyle
end
Wernicke --> pSTS
Amygdala --> pSTS
Spt --> pSTS
end
subgraph Executive[" EXECUTIVE PROCESSING "]
direction LR
PFC[PFC<br/>LLM]:::execStyle
CorpCall[Corp<br/>Call]:::execStyle
ACC[ACC]:::execStyle
Salience[Salience<br/>Network]:::execStyle
PFC --> CorpCall --> ACC --> Salience
end
subgraph Subcortical[" Subcortical Loop "]
direction LR
BG[BG]:::subcortStyle
Cereb[Cereb]:::subcortStyle
WorkingMem[Working<br/>Memory]:::subcortStyle
VTA_DA[VTA DA]:::subcortStyle
BG --> Cereb
Cereb --> WorkingMem
WorkingMem --> VTA_DA
end
subgraph Phase7[" Phase 7: SPEECH PRODUCTION <br/><i>Levelt 1989</i> "]
direction LR
BrocaLLM[Broca LLM<br/>production]:::speechStyle
MotorCortex[Motor Cortex<br/>M1]:::speechStyle
BrocaLLM --> MotorCortex
end
subgraph Phase6[" Phase 6: RETRIEVAL "]
direction LR
RetrievalEngine[Retrieval<br/>Engine]:::retrievalStyle
ProceduralCache[Procedural<br/>Cache]:::retrievalStyle
RetrievalEngine --> ProceduralCache
end
subgraph Neuromod[" NEUROMODULATOR SYSTEM "]
direction LR
VTA[VTA DA]:::neuromodStyle
Hypothalamus[Hypothalamus<br/>Homeostasis]:::neuromodStyle
LC[LC<br/>NE]:::neuromodStyle
Raphe[Raphe Nuclei<br/>5-HT]:::neuromodStyle
NucleusBasalis[Nucleus Basalis<br/>ACh]:::neuromodStyle
end
MemorySystem[/MEMORY SYSTEM<br/>6-layer CLS<br/><i>McClelland 1995</i>/]:::memoryStyle
UserInput --> Thalamus
SensoryBuffer --> Wernicke
SensoryBuffer --> Amygdala
SensoryBuffer --> Spt
pSTS --> Salience
Salience --> PFC
ACC --> BG
VTA_DA --> PFC
PFC --> BrocaLLM
Phase6 --> Executive
Neuromod -.modulates.-> Executive
Neuromod -.modulates.-> Phase23
MemorySystem -.stores.-> Phase6
Neuromod --> MemorySystem
classDef inputStyle fill:#F5E8D3,stroke:#B8A078,stroke-width:2px,color:#5C5847
classDef sensoryStyle fill:#E8DFF0,stroke:#B8A5C9,stroke-width:2px,color:#5C4A6B
classDef ventralStyle fill:#DDE5D0,stroke:#A8B590,stroke-width:2px,color:#4A5238
classDef dorsalStyle fill:#E0D5E8,stroke:#B09BC4,stroke-width:2px,color:#5C4A6B
classDef mergeStyle fill:#F0E0D8,stroke:#C4A898,stroke-width:3px,color:#5C3E2E
classDef execStyle fill:#E5DCE8,stroke:#B09BC4,stroke-width:2px,color:#4A3858
classDef subcortStyle fill:#F0EAD8,stroke:#B8AC80,stroke-width:2px,color:#5C5030
classDef speechStyle fill:#DDE5D0,stroke:#A8B590,stroke-width:2px,color:#4A5238
classDef retrievalStyle fill:#E8DFF0,stroke:#B8A5C9,stroke-width:2px,color:#5C4A6B
classDef neuromodStyle fill:#F5E8D8,stroke:#C4B088,stroke-width:2px,color:#5C4A30
classDef memoryStyle fill:#EADFEC,stroke:#B8A0C0,stroke-width:3px,color:#4A3858
23 regions across 10 lobes with anatomically correct hemisphere assignments. Six regions use LLM calls (Wernicke, Amygdala R+L, PFC, Broca, Visual Cortex); all others are algorithmic.
| Region | Hemisphere | Function |
|---|---|---|
| Prefrontal Cortex (PFC) | Bilateral | LLM reasoning, goal tree, entity extraction |
| ACC | Bilateral | Conflict monitoring, error accumulation |
| Broca's Area | Left | LLM language production |
| Thalamus | Bilateral | Sensory relay and gating |
| Hypothalamus | Bilateral | Homeostatic regulation |
| Amygdala | R/L split | R=fast appraisal, L=contextual evaluation |
| Wernicke's Area | Left | LLM semantic analysis |
| Auditory Cortex | L + R | Speech (L) + prosody (R) |
| Visual Cortex | Bilateral | Image processing |
| Angular Gyrus | Left | Cross-modal semantic binding |
| pSTS | Left | Multisensory stream merging |
| Spt | Left | Auditory-motor interface |
| Motor Cortex | Left | Final output execution |
| Salience Network | Bilateral | DMN/ECN/Creative mode switching |
| Basal Ganglia | Bilateral | Go/NoGo action selection |
| Corpus Callosum | Bilateral | Inter-hemisphere integration |
| Cerebellum | Bilateral | Forward model prediction |
| VTA | Bilateral | Dopamine, reward prediction error |
| Brainstem | Bilateral | Arousal regulation |
| mPFC | Bilateral | Self-referential processing |
| TPJ | Right | Theory of Mind |
| Insula | Bilateral | Interoceptive awareness |
| Hippocampus | Bilateral | Fast encoding, modality tagging |
Six-layer pipeline: Atkinson-Shiffrin + CLS (McClelland 1995) + Baddeley working memory.
Sensory Buffer --> Working Memory --> Hippocampal Staging --> Episodic Store
| |
| Consolidation
| |
+----------> Semantic Store
Procedural Store
| Layer | Key Mechanism |
|---|---|
| Sensory Buffer | Per-cycle flush (Sperling 1960) |
| Working Memory | Baddeley model: phonological + visuospatial + episodic buffer |
| Hippocampal Staging | ACh-modulated fast encoding |
| Episodic Store | Ebbinghaus forgetting, reconsolidation |
| Semantic Store | Knowledge graph with confidence tagging, Leiden community detection, spreading activation |
| Procedural Store | DA-gated learning, 3-stage skill acquisition (Fitts 1967) |
The semantic store includes a graph analysis layer inspired by connectomics research. The knowledge graph is not a flat triple store — it has structure.
| Feature | Mechanism | Neuroscience |
|---|---|---|
| Community Detection | Leiden algorithm on concept graph | Cortical columns (Mountcastle 1997) |
| Hub Concepts | Degree-ranked central nodes | Rich-club organization (van den Heuvel & Sporns 2011) |
| Surprising Connections | Cross-community bridge scoring | Long-range cortical projections |
| Confidence Tagging | EXTRACTED / INFERRED / AMBIGUOUS per edge | Signal Detection Theory (Green & Swets 1966) |
| Graph Diff | LTP (new) / LTD (lost) / pruning classification | Synaptic plasticity (Bliss & Lomo 1973) |
| Compressed Context | Graph summary instead of raw memory dump | Chunking (Miller 1956) |
| Embedding Cache | SHA256 content-addressable LRU | Long-term potentiation (faster reactivation) |
| Cell Assemblies | Hyperedge groups (3+ concepts) with co-activation | Hebb's Cell Assembly (1949) |
| Assembly Co-activation | Active member triggers ensemble spread | Neural ensemble synchronization |
| Graph Pruning | Weight decay + threshold pruning during consolidation | Synaptic pruning (Huttenlocher 1979) |
| Metacognitive Query | MCP tools for self-inspecting knowledge | Metacognition (Flavell 1979) |
| Community-Aware Activation | Intra-community spread bonus in retrieval | Cortical column facilitation |
Confidence flows into the neuromodulator system: AMBIGUOUS edges raise NE (alertness) and ACh (learning), triggering ACC conflict monitoring. EXTRACTED edges pass through without friction. This mirrors how the brain allocates more attention to uncertain information.
Cell assemblies (hyperedges) enable group-level memory: when one member of an assembly activates during retrieval, all members receive co-activation spread — just as Hebbian ensembles fire as coordinated units. The MCP knowledge server exposes query_graph, get_neighbors, list_communities, find_hubs, find_bridges, and get_assemblies as agent-callable tools, enabling metacognitive self-inspection.
A workspace-aware curation layer sits on top of the 6-layer memory system. It turns CBA from a general conversational agent into a business-logic curator for coding agents: you feed in specs, decisions, PDFs, or ad-hoc chat, and the system stores them losslessly, partitions them by project, detects contradictions, asks when something's ambiguous, and exposes the curated context for downstream tools.
Status: Phase 0-8 are implemented and wired into the runtime, memory layer, extraction pipeline, dashboard backend, and React dashboard. The local-only plan set in docs/superpowers/plans/2026-04-17-phase-*.md remains the detailed TDD history; the public README now reflects the delivered behavior.
| Phase | Delivered surface | Runtime result |
|---|---|---|
| 0 - Workspace & ontology foundation | WorkspaceStore, OntologyStore, universal seed, migration runner |
Every session has a current workspace, a personal workspace exists by default, and ontology types can be resolved per workspace plus __universal__. |
| 1 - Raw Vault & schema enrichment | SHA256 RawVault, workspace/provenance/importance/decay columns across memory stores |
Original inputs are preserved or pointer-tracked, deduplicated by hash, and linked back from graph/staging/episode records. |
| 2 - Contradictions & open questions | ContradictionsStore, OpenQuestionsStore, severity rules, batch subject lookup |
Ambiguity and conflicts are durable queues instead of transient log messages; severe items can block the response path. |
| 3 - Multi-stage extractor | Triage, Extract, Temporal Resolve, Validate, Severity, Refine, Orchestrator | The old single PSC call is replaced with workspace-aware, ontology-constrained extraction that writes only to staging and curation queues. |
| 4 - Personal adapter | PersonalAdapter over legacy identity_facts |
Personal memory keeps backward compatibility while exposing user/agent facts as workspace-style Person nodes. |
| 5 - Pipeline integration | Orchestrator front-door wiring, response_mode, retrieval contradiction/gap metadata, Wernicke workspace hints, Broca block formatting |
Normal answers, appended clarification questions, and block-mode clarification responses all flow through the neural pipeline without changing the 7-phase topology. |
| 6 - Decay policy | Workspace/type/edge decay policy, importance_score, never_decay, all-workspaces dreaming |
Business-critical facts can be protected from forgetting while personal and low-importance memories still decay normally. |
| 7 - Domain templates | software-project, research-notes, personal-knowledge, apply/upgrade/downgrade APIs |
New workspaces can adopt a ready ontology, minor upgrades can apply safely, major upgrades require confirmation, and downgrades are refused. |
| 8 - Visualization & human-in-the-loop | Workspace APIs, curation APIs, source/timeline/export/model APIs, dashboard selectors and inboxes | The dashboard can switch workspaces, inspect scoped graphs, resolve questions/contradictions/proposals, preview raw sources, export curated context, and select per-stage models. |
| Feature | Purpose | Neuroscience anchor |
|---|---|---|
| Multi-workspace knowledge graph | Separate personal, billing-service, research-notes etc. with optional cross-references |
Bartlett (1932) schema theory; van Kesteren et al. (2012) schema-dependent encoding |
| Raw Vault | SHA256-addressed lossless storage of every input (text, image, PDF, audio). Small files copied, large files pointer-tracked | Johnson, Hashtroudi & Lindsay (1993) source monitoring |
| 4-tier confidence ontology | PROVISIONAL → STABLE → CANONICAL → USER_GROUND_TRUTH per node/relation type, auto-promoted on re-occurrence |
Kadavath et al. (2022) LLM self-confidence miscalibration |
| Multi-stage extraction | Replaces the single-call PSC with 6 stages: Triage → Extract (ontology-constrained) → Temporal Resolve → Validate → Severity Branch → Broca Refine | McClelland et al. (1995) Complementary Learning Systems |
| Temporal Resolve (Stage 2.5) | Distinguishes state changes ("지금은 Go로 바꿨어") from genuine contradictions — prevents false-positive clarification blocks | Conway (2005) time-indexed self-memory |
| Severity-tiered clarification | Ambiguity/contradictions become first-class pipeline outputs: block (severe → respond with question), append (moderate → answer + question), normal |
Botvinick et al. (2001) ACC conflict monitoring |
| Contradictions + Open Questions stores | Persistent human-in-the-loop queues. Contradictions carry both sides' source snippets; open questions track unanswered clarifications | Hart (1965) feeling-of-knowing |
| Pattern separation for Events | Similar events with nearby timestamps trigger a merge-or-distinct clarification instead of silent collapse | Yassa & Stark (2011) dentate gyrus |
never_decay + importance_score |
Business logic / specs / decisions can be protected from normal forgetting; emphasis words and reinforcement modulate decay | LeDoux (1996) amygdala event-level modulation |
| Domain templates | Drop-in ontologies: software-project (Requirement, Decision, Module, Interface, Constraint, Risk…), research-notes, personal-knowledge |
Ashby & Maddox (2011) category learning |
| Coding agent export preview | Filterable JSON export (by confidence tier, importance, never_decay) that matches the MCP-compatible response shape |
— |
User input (text/image/audio/PDF)
│
├─ Raw Vault (SHA256 dedup + integrity)
│
├─ Stage 1 Triage → workspace routing, multi-label input kind
├─ Stage 2 Extract → ontology-constrained structured output
├─ Stage 2.5 Temporal Resolve → supersede / reinforce / contradict branch
├─ Stage 3 Validate → contradiction + missing-premise detection
├─ Stage 4 Severity Branch → normal / append / block
└─ Stage 5 Broca Refine → personal workspace only
│
└─ Persist (staging-only) → ConsolidationEngine promotes to semantic/episodic
Writes land in hippocampal staging only — semantic and episodic promotions happen through the existing ConsolidationEngine, preserving the CLS fast/slow distinction.
- Workspace selector in the HUD, current-workspace badge, workspace-scoped graph requests, and optional cross-reference edges.
- Curation Inbox with Open Questions, Contradictions, and Ontology Proposals tabs. Actions taken in chat or Inbox emit WebSocket events so both surfaces stay in sync.
- Raw Vault drill-down from source-linked nodes and edges back to metadata, extracted text, or raw bytes.
- Timeline view for temporal supersede chains, so "old state -> new state" updates are inspectable instead of flattened.
- Export Preview modal with confidence, importance,
never_decay, and raw-vault filters before a coding agent consumes context. - Model Selector backed by LiteLLM provider inventory. Triage, extract, temporal classify, and refine stages are configurable independently with opaque model identifiers.
Six neurochemical systems with different decay rates and anatomically correct source nuclei.
| NT | Source | Effect | Decay |
|---|---|---|---|
| DA | VTA/SNc | Reward prediction error | 0.85 |
| NE | Locus Coeruleus | Urgency, alertness | 0.85 |
| 5-HT | Dorsal Raphe | Patience, inhibition | 0.90 |
| ACh | Nucleus Basalis | Learning strength | 0.85 |
| CORT | HPA Axis | Stress response | 0.93 |
| EPI | Adrenal Medulla | Fight-or-flight | 0.75 |
Real-time 3D brain visualization and knowledge-layer curation: React 19 + Three.js + Zustand + WebSocket.
brain-agent dashboard --port 3000- 23 brain regions with activation glow and sequential cascade
- Signal particles flowing between regions
- 25+ anatomical neural connections
- HUD with network mode, 6 neurotransmitter bars, current workspace, inbox count, export controls, and model controls
- Memory flow pipeline with live counts
- Knowledge graph visualization with workspace filters, optional cross-reference edges, community coloring, hub highlighting, confidence overlays, and Raw Vault source preview
- Curation Inbox for answering open questions, resolving contradictions, and approving or rejecting ontology proposals
- Timeline view for superseded temporal facts
- Export Preview for MCP-compatible coding-agent context
- Per-stage model selector sourced from
/api/llm/providers - Audio input with voice mode
- Multimodal input (image, audio, text)
CBA/
|-- brain_agent/
| |-- agent.py # Main entry point
| |-- pipeline.py # 7-phase neural pipeline with extraction-orchestrator integration
| |-- config/ # Pydantic configuration, including extraction/workspace settings
| |-- core/ # Signals, neuromodulators, router, workspace primitives
| |-- regions/ # 23 brain regions
| |-- memory/ # CLS stores plus workspace, ontology, raw vault, curation, templates, decay
| |-- extraction/ # Multi-stage extractor: triage, extract, temporal, validate, severity, refine
| |-- migrations/ # Schema migration runner and knowledge-layer migrations
| |-- providers/ # LLMProvider abstraction and LiteLLM implementation
| |-- dashboard/ # FastAPI app, event emitter, provider inventory, routers
| |-- tools/ # Tool registry
| |-- mcp/ # MCP integration
| `-- middleware/ # Middleware chains
|-- dashboard/ # React + Three.js + Zustand dashboard
|-- tests/ # Python, dashboard API, extraction, memory, and frontend tests
|-- assets/ # README images
|-- .env.example # Environment template
`-- LICENSE # MIT
The 2026-04-17 knowledge-layer rollout is complete across Phase 0-8. The local-only TDD plans remain under docs/superpowers/plans/; this table summarizes the public state.
| Phase | Scope | Status |
|---|---|---|
| 0 - Foundation | WorkspaceStore, OntologyStore, universal seed, migration runner |
Complete |
| 1 - Raw Vault & Schema Enrichment | SHA256 raw vault, workspace/provenance/importance/decay columns, ChromaDB workspace metadata | Complete |
| 2 - Contradictions & Open Questions | Severity-tiered durable curation stores and batch lookup | Complete |
| 3 - Multi-stage Extractor | Triage -> Extract -> Temporal Resolve -> Validate -> Severity -> Refine orchestrator | Complete |
| 4 - Personal Adapter | Backward-compatible identity_facts bridge to workspace nodes |
Complete |
| 5 - Pipeline Integration | Orchestrator replaces PSC, response modes, retrieval S1/S2, expression-mode wiring | Complete |
| 6 - Decay Policy | Workspace/type/edge decay, importance_score, never_decay, all-workspaces dreaming |
Complete |
| 7 - Domain Templates | Software project, research notes, personal knowledge templates with upgrade semantics | Complete |
| 8 - Visualization & Human-in-the-Loop | Dashboard workspace, curation, source, timeline, export, and model-selection surfaces | Complete |
pytest # Python suite
pytest --cov # Python coverage
cd dashboard && npm test # React/Vitest dashboard suite| Branch | Description |
|---|---|
main |
Stable release with Knowledge Layer Phase 0-8 implemented. |
graphify |
Knowledge graph analysis: Leiden clustering, cell assemblies, MCP metacognition, dashboard viz |
openclaw |
Extended features: MCP, tool system, middleware |
This framework is grounded in 50+ published neuroscience papers spanning 1929–2023.
| Citation | Topic | Region |
|---|---|---|
| Hubel & Wiesel (1959) | Receptive fields in visual cortex | Visual Cortex (V1) |
| Milner (1971) | Hippocampal hemisphere specialization | Hippocampus |
| Ungerleider & Mishkin (1982) | Two cortical visual systems (ventral/dorsal) | Visual Cortex |
| Baars (1988) | Global Workspace Theory — broadcast mechanism | Pipeline |
| Levelt (1989) | Speaking: From Intention to Articulation | Motor Cortex, Broca |
| Mink (1996) | Basal ganglia Go/NoGo gating | Basal Ganglia |
| LeDoux (1996) | The Emotional Brain | Amygdala |
| Morris et al. (1998) | Right hemisphere automatic emotional processing | Amygdala R |
| Baddeley (2000) | Working memory: episodic buffer and capacity limits | Working Memory |
| Calvert et al. (2000) | pSTS superadditivity for congruent stimuli | pSTS |
| Wheeler et al. (2000) | Multisensory memory retrieval reactivation | pSTS |
| Eichenbaum (2000) | Hippocampus and entity extraction | PFC |
| Goldberg (2001) | PFC lateralization: left=routine, right=novel | PFC |
| Botvinick et al. (2001) | Conflict monitoring and cognitive control | ACC |
| Holroyd & Coles (2002) | Error-related negativity | ACC |
| Corbetta & Shulman (2002) | Dorsal/ventral attention streams | Attention |
| Saxe & Kanwisher (2003) | People thinking about thinking people | TPJ |
| Glascher & Adolphs (2003) | Amygdala response processing | Amygdala |
| Hickok et al. (2003) | Speech production planning | Spt |
| Beauchamp et al. (2004) | Audiovisual integration in pSTS | pSTS |
| Critchley et al. (2004) | Interoceptive awareness | Insula |
| Squire (2004) | Hippocampal memory binding | Hippocampus |
| Beeman (2005) | Right hemisphere creative insight | PFC |
| D'Argembeau et al. (2005) | Self-referential processing in mPFC | mPFC |
| Frank (2005) | Direct/indirect pathway balance | Basal Ganglia |
| Frith & Frith (2006) | Neural basis of mentalizing | TPJ |
| Northoff et al. (2006) | Self-referential processing in mPFC | mPFC |
| Guenther (2006) | DIVA model of speech production | Motor Cortex |
| Sherman & Guillery (2006) | Exploring the Thalamus | Thalamus |
| Paulus & Stein (2006) | Interoception and risk processing | Insula |
| Barrett (2006) | Constructionist emotion theory | Amygdala |
| Hickok & Poeppel (2007) | Dual-stream model of speech processing | Wernicke, Broca, Spt, A1 |
| Sherman (2007) | Thalamus is more than just a relay | Thalamus |
| Aron (2007) | Conflict-induced braking (GABA) | PFC |
| McAlonan et al. (2008) | Thalamic reticular nucleus attention gating | Thalamus |
| Ito (2008) | Cerebellar forward models and motor learning | Cerebellum |
| Graybiel (2008) | Procedural memory pattern caching | Procedural Store |
| Pessoa (2008) | Content-driven dynamic activation | Pipeline |
| Van Overwalle (2009) | Social cognition and TPJ meta-analysis | TPJ |
| Craig (2009) | How Do You Feel — Now? Interoception | Insula |
| Singer et al. (2009) | Emotion-interoception integration | Insula |
| Dehaene (2009) | Orthographic visual processing (LGN) | Thalamus |
| Price (2010) | Reading and the angular gyrus | Angular Gyrus |
| Buchsbaum et al. (2011) | Verbal working memory | Spt |
| Menon (2011) | Network mode detection and switching | Salience Network |
| Ramachandran (2011) | Cross-modal abstraction | Angular Gyrus |
| Isaacson & Scanziani (2011) | E/I balance compensation (GABA) | Pipeline |
| Fleming & Dolan (2012) | Neural basis of metacognitive ability | PFC |
| Yeung & Summerfield (2012) | Metacognition in decision-making | PFC |
| Ghosh & Gilboa (2014) | Schemas always active in mPFC | mPFC |
| Buzsáki (2015) | Hippocampal sharp-wave ripples | Consolidation |
| Beaty et al. (2018) | Creative cognition and the default mode network | Salience Network |
| Citation | Topic | System |
|---|---|---|
| Sperling (1960) | Sensory buffer iconic memory | Sensory Buffer |
| Fitts (1967) | Three-stage skill acquisition | Procedural Store |
| Anderson (1994) | Retrieval-induced forgetting | Retrieval Engine |
| McClelland et al. (1995) | Complementary Learning Systems | Consolidation |
| Wozniak (1990) | SM-2 spaced repetition algorithm | Episodic Store |
| Nader (2000) | Memory reconsolidation | Episodic Store |
| Yassa & Stark (2011) | Pattern separation in dentate gyrus | Hippocampal Staging |
| Winocur & Moscovitch (2011) | Episodic → semantic transformation | Consolidation |
| Tononi & Cirelli (2006) | Synaptic homeostasis hypothesis | Homeostatic Scaling |
| Zielinski et al. (2018) | Slow-wave sleep consolidation | Consolidation |
| Park et al. (2023) | Generative Agents: reflection mechanism | Reflection |
| Diekelmann & Born (2010) | Memory consolidation during sleep | Dreaming Engine |
| Rasch & Born (2013) | About sleep's role in memory | Dreaming Engine |
| Citation | Topic | System |
|---|---|---|
| Cannon (1929) | Fight-or-flight response | Epinephrine |
| Gold & Van Buskirk (1975) | Epinephrine enhances memory | Epinephrine |
| Schultz (1997) | Dopamine reward prediction error | Dopamine / VTA |
| Cahill & McGaugh (1998) | Emotion and memory consolidation | Epinephrine |
| Grace (2000) | Tonic vs phasic dopamine firing | Dopamine / VTA |
| de Quervain et al. (2000) | Cortisol impairs memory retrieval | Cortisol |
| Doya (2002) | Serotonin and temporal discounting | Serotonin |
| Sapolsky (2004) | Stress and cortisol effects on cognition | Cortisol |
| Dickerson & Kemeny (2004) | Social-evaluative threat and cortisol | Cortisol |
| Aston-Jones & Cohen (2005) | Adaptive gain theory (norepinephrine) | Norepinephrine |
| Phelps & LeDoux (2005) | Amygdala-cortisol interaction | Amygdala |
| Friston (2005) | Predictive coding framework | Pipeline |
| Hasselmo (2006) | ACh gating: novelty, learning, plasticity | Acetylcholine |
| Buzsáki (2006) | Cortical oscillations and GABA | GABA |
| McEwen (2007) | Allostatic load and stress persistence | Cortisol |
| Kirschbaum et al. (1995) | Cortisol and stress recovery | Cortisol |
| Schneider & Shiffrin (1977) | Automatic vs controlled processing | Pipeline |
| Lamme (2006) | Recurrent processing and consciousness | Pipeline |
| Rolls (2013) | Pattern completion in CA3 | Retrieval Engine |
| Citation | Topic | System |
|---|---|---|
| Green & Swets (1966) | Signal Detection Theory | Confidence Scoring |
| Miller (1956) | Chunking and working memory capacity | Compressed Context |
| Bliss & Lomo (1973) | Long-term potentiation | Graph Diff (LTP) |
| Huttenlocher (1979) | Synaptic pruning during development | Graph Diff (pruning) |
| Mountcastle (1997) | Cortical column modularity | Leiden Community Detection |
| Watts & Strogatz (1998) | Small-world network topology | Knowledge Graph |
| van den Heuvel & Sporns (2011) | Rich-club organization in brain networks | Hub Concept Detection |
| Frankland & Bontempi (2005) | Systems consolidation | Leiden-based Consolidation |
| Reyna & Brainerd (1995) | Fuzzy-trace theory (gist extraction) | Compressed Context |
| Hebb (1949) | Cell Assembly theory | Hyperedges / Co-activation |
| Flavell (1979) | Metacognition | MCP Knowledge Server |
| Collins & Loftus (1975) | Spreading activation | Community-aware retrieval |
| Citation | Topic | System |
|---|---|---|
| Bartlett (1932) | Schema theory: recall as reconstruction | Workspace as schema frame |
| Hart (1965) | Feeling-of-knowing phenomenon | Open Questions store |
| Brown & McNeill (1966) | Tip-of-the-tongue | Expression-mode gap detection |
| Johnson, Hashtroudi & Lindsay (1993) | Source monitoring framework | Raw Vault, epistemic source tagging |
| Moscovitch & Nadel (1997) | Multiple Trace Theory | Append-only versioning, supersedes |
| Squire (1992) | Multiple memory systems | Staging-only write discipline |
| Eichenbaum (2000) | Source binding | Knowledge graph provenance |
| Conway (2005) | Memory and the Self — time-indexed facts | Stage 2.5 Temporal Resolve |
| Tse et al. (2007) | Schema-dependent consolidation | Workspace-aware encoding |
| Ashby & Maddox (2011) | Category learning | Ontology type hierarchy |
| van Kesteren et al. (2012) | Schema-dependent encoding | Multi-workspace routing |
| Ghosh & Gilboa (2014) | Schemas always active | Session workspace persistence |
| Kadavath et al. (2022) | LLM self-confidence miscalibration | 4-tier confidence (PROVISIONAL → STABLE → CANONICAL → USER_GROUND_TRUTH) |