Documentation Map
August 17, 2026 · View on GitHub
Start here. Everything under
docs/organized by what you're looking for.
| Looking for… | Go to |
|---|---|
| What this project is, quick start, benchmarks | README.md |
| How to activate the MCP tools in your agent | CLAUDE.md |
| Version history | CHANGELOG.md |
| Where the project is headed | roadmap.md |
Layout
docs/
├── design/ Architecture & algorithm design docs
├── benchmarks/ Benchmark harnesses, protocols & results
├── evaluations/ Audit reports, experiment logs, optimization plans
├── paper/ Paper drafts, outline, review & LaTeX source
└── research/ Literature survey (notes, references.bib, 中文翻译)
design/ — how it works
| Doc | What it covers |
|---|---|
| design.md | Original founding design: the causal-layer beachhead diagnosis |
| complete-memory-system.md | 完整记忆系统架构 — one graph, one engine, one loop (the full system) |
| unified-memory-design.md | 三层统一记忆架构 — facts + causal + unified retrieval (RRF) |
| architecture.md | Component-level architecture (MCP server, store, engine) |
| algorithm-design.md | Algorithm formalization: gated propagation + online learning + graph sparsification |
| hippocampus-design.md | Hippocampus engine: CSR spreading activation, DG/CA3/CA1, SWR |
| refutation-design.md | Causal-edge refutation: real-time confidence scoring (A/B/C/D/F) for LLM-extracted edges |
| multi-hop-expansion.md | Multi-hop / open-domain retrieval via graph expansion |
benchmarks/ — measuring it
| Doc | What it covers |
|---|---|
| causal-eval-2026.md | CausalEval — our graph-grounded causal memory benchmark (primary) |
| locomo.md | LoCoMo fact-recall results & optimization matrix |
| longmemeval.md | LongMemEval results |
| memora.md | Memora weekly persona results |
| tau2-airline.md | τ²-bench airline behavioral A/B |
| amc-2026.md | Agent Memory Challenge (AMC/01) submission details |
evaluations/ — what we found
| Doc | What it covers |
|---|---|
| edge-accuracy-audit.md | Causal-edge labeling accuracy audit (83% agreement, 0 severe) |
| performance-audit-2026-08.md | Performance & precision audit |
| optimization-plan-2026-08.md | 2026-08 optimization plan |
| code-review-retrieval.md | Retrieval & write-path code review (2026-08-01) |
| p8-p9-experiments.md | P8/P9 实验设计 |
| mem0-eval-alignment.md | mem0 评测对齐设计与实施 |
| mem0-eval-followup.md | mem0 对齐后续任务 (F1-F4) |
| mem0-eval-final.md | mem0 对齐收尾 (F5-F8) |
| insights-didi-memory-2026.md | DiDi Darwinian Memory + UltraHorizon research notes |
paper/ — writing it up
| Doc | What it covers |
|---|---|
| paper-outline.md | Paper outline |
| paper-full-draft.md | Full draft |
| paper-section4-experiments.md | Section 4 (experiments) |
| paper-review.md | Nature-style reviewer assessment |
| paper-latex/ | LaTeX source |
research/ — background
Literature survey across causal inference, neuroscience, cognitive psychology, and computational AI. See research/README.md for the index, design-lineage.md for how the papers map to design decisions, and references.bib for the bibliography. 中文翻译在 zh/。
Project-level research backdrop: research-backdrop.md (中文).