Truth
records
Evidence
justifies
Signals
measures
Narrative
proposes
Knowledge
preserves
XMN

EVIDENCE-BASED COGNITIVE ARCHITECTURE

Observe.
Measure.
Understand.

// Never pretend certainty.

An evidence-based cognitive architecture for AI.
Every conclusion is traceable. Every interpretation is revisable. XMN은 대화를 기억합니다. 하지만 확신을 가장하지 않습니다.
모든 결론은 추적 가능하고, 모든 해석은 수정 가능합니다. XMNは会話を記憶しますが、確信を装いません。
すべての結論は追跡可能であり、すべての解釈は修正可能です。 XMN记住对话，但从不假装确定。
每个结论都可追溯，每个解释都可修正。

Try Playground 플레이그라운드 試してみる 立即体验 Read Constitution 헌법 읽기 憲法を読む 阅读宪法

// STEP 1 · WHY DIFFERENT

Why is XMN different? 단순 저장이 아닙니다. 이해입니다. ただの記憶ではなく、理解です。 不只是存储，而是理解。

TYPICAL AI

Input
↓
Answer

No trace. No evidence.
No explanation.

❌ Why? — cannot explain.
VS

XMN

Input
↓
Truth
↓
Evidence
↓
Signals
↓
Interpretation
↓
Knowledge
✅ Fully traceable. Evidence versioned.

// STEP 2 · EXPERIENCE IT YOURSELF

Try it now — paste any text. XMN을 직접 경험하세요 XMNを体験する 体验XMN

Paste any conversation. Watch XMN explain not only what it remembers, but why. 어떤 대화든 붙여넣으세요. XMN이 무엇을 기억하는지, 그리고 왜 그렇게 이해했는지를 보여줍니다. 会話を貼り付けてください。XMNが何を記憶するかだけでなく、なぜそう理解したかを説明します。 粘贴任何对话。XMN将解释它记住了什么，以及为什么这样理解。

30-second interactive demo
Phrase Concept Cognitive Semantic Unmapped

CONCEPT GRAPH

PIPELINE

✓
Particle strip
Korean/Japanese particles stripped
↓
✓
Phrase
career anxiety → career_anxiety
↓
✓
Concept map
family → relationship
↓
✓
Cognitive signal
avoiding → coping_pattern
↓
○
Semantic
skipped
↓
✓
Narrative
generating...
RECOGNITION COVERAGE —
Phrase0%
Concept0%
Cognitive0%
Semantic0%
Unmapped0%

CONCEPTS

career_anxiety stress family avoidance

NARRATIVE

Career · Relationship
Career challenges — constrained by family responsibilities — holding off on change.
causecareer pressure
constraintfamily responsibilities
directionavoiding change
Evidence · Narrative-v1 · 2026.06

// WHAT JUST HAPPENED?

XMN didn't just extract keywords.
It moved through five distinct cognitive layers.
Each layer has a different role. None can be skipped.

// STEP 3 · HOW XMN THINKS

The five layers of understanding XMN의 사고 방식 XMNの思考方法 XMN的思维方式

Five distinct layers. Each with a different role. Never conflated. 다섯 개의 독립 계층. 각자 다른 역할. 절대 혼용되지 않습니다. 5つの独立した層。それぞれ異なる役割。決して混同されません。 五个独立层次，各有不同职责，从不混淆。

Truth
records
Evidence
justifies
Signals
measures
Interpretation
proposes
Knowledge
preserves

Each layer has
a distinct verb.
The verbs matter.

Truth

Memory Graph

Observed facts and relationships. Nodes, edges, weights. Never rewritten — only accumulated. The ground truth XMN never contradicts.

Truth is immutable. Memory Graph records. Never overwrites.
Evidence

Reproducible Chain

Every interpretation traces to explicit evidence, versioned by algorithm and concept map. In 2030, XMN can reproduce why it thought what it thought in 2026.

Evidence connects Truth to Interpretation. Always versioned.
Behavioral Signals

Sensor Layer

Language patterns converted to measurements — not inferences. An agency score is a reading, like temperature. XMN never labels a person from a signal.

Signals are measured, not inferred. Signals describe patterns, not people.
Interpretation

Narrative Compression

Hypotheses generated from evidence. Always revisable. Reinterpretation doesn't change memory — it changes understanding.

Interpretation is revisable. It is a hypothesis, not a fact.
Knowledge

Life Chapter

What has remained meaningful across time. Not produced by overwriting — accumulated through repeated reinterpretation.

Knowledge accumulates. Never produced by overwriting history.
// STEP 4 · THE PRINCIPLES BEHIND EVERY DECISION The principles XMN will never violate Read the Constitution ↓

These layers don't exist by accident. They are defined by a set of principles that govern every design decision in XMN.

I

Truth is immutable.

Memory Graph records observed facts. Truth is never rewritten.

II

Evidence is reproducible.

Every interpretation must be traceable to explicit evidence and reproducible with the same algorithm version.

III

Interpretation is revisable.

Interpretations are hypotheses, not facts. They may evolve as new evidence becomes available.

IV

Knowledge accumulates.

Knowledge emerges from the accumulation of interpretations over time. Never by overwriting history.

V

Reinterpretation does not change memory. It changes understanding.

Memory remains intact. Only the interpretation of memory evolves.

VI

XMN does not judge.

XMN records. XMN detects. XMN signals. The moment XMN becomes a coach, the constitution is broken.

VII

Observation precedes interpretation.

Signals are measured, not inferred. Signals describe patterns, not people. XMN does not diagnose. XMN observes.

// Design Rule

When evaluating any new feature, ask one question first:

"Does this violate the XMN Constitution?"

If the answer is yes, the design must be reconsidered before implementation.

// STEP 6 · EVOLUTION

How we got here 베일드노트에서 XMN으로 VeiledNoteからXMNへ 从VeiledNote到XMN

FOUNDATION
VeiledNote Voice Engine ✓
The origin — voice-to-memory for AI sessions
MEMORY
Memory Graph ✓
Node · Edge · MPI · Lifecycle · 4-layer memory
EXPLAINABILITY
Evidence Architecture ✓
Trace · Evidence · Algorithm versioning · Full reproducibility
PIPELINE
Semantic Pipeline ✓
Phrase → Concept → Alias → Semantic · Korean/English
BEHAVIORAL
Behavioral Signals Layer ✓
Agency · Coping · Attention · Unresolved Loop · Stability
NARRATIVE
Narrative Compression ✓
Theme · Goal · Constraint · Direction · Evidence snapshot
KNOWLEDGE
Life Chapter △ in progress
10 Narratives → Knowledge accumulation over time
MEANING
Meaning Trajectory future
Same event — meaning changes across time
PLATFORM
SDK Release future
pip install xmn · REST API · Python · Enterprise

// STEP 5 · RESEARCH & DESIGN PHILOSOPHY

The thinking behind the architecture 연구 및 설계 철학 研究と設計哲学 研究与设计哲学

XMN sits at the intersection of cognitive science, AI memory, and explainability. This is where we document the thinking. XMN은 인지과학, AI 기억, 설명가능성의 교차점에 있습니다. 여기에 그 사고 과정을 기록합니다. XMNは認知科学、AIメモリ、説明可能性の交差点にあります。 XMN处于认知科学、AI记忆和可解释性的交叉点。

ARCHITECTURE · Working Paper 001

The Five-Layer Cognitive Model

Why Truth, Evidence, Signals, Interpretation, and Knowledge must remain distinct layers — and what breaks when they are conflated. Each layer has one verb. The verbs matter.

XMN Constitution · Phase 16 · 2026.06 · published
EPISTEMOLOGY · Working Paper 002

XMN Epistemology

Truth records. Evidence justifies. Signals measure. Interpretations propose. Knowledge preserves. The same structure as scientific method: Observation → Evidence → Hypothesis. Interpretation is always a hypothesis.

XMN Constitution · 2026.06 · published
BEHAVIORAL SIGNALS · Working Paper 003

Signals Are Measured, Not Inferred

agency = low is not a personality diagnosis. It is a sensor reading — like temperature. XMN stores the observation, never the label. The moment it labels a person, the constitution is broken.

Phase 17 · 2026.06 · published
EXPLAINABILITY · Working Paper 004

Evidence Versioning — Reproducible in 2030

Every Narrative stores {algorithm, pipeline, semantic_adapter, concept_map} version. A Narrative produced in 2026 must be reproducible in 2030 with the same inputs. This is what separates XMN from black-box memory.

Phase 15 · 2026.06 · published
NARRATIVE · Working Paper 005

People Remember Stories, Not Keywords

"stress, career, family" — three concepts. But what a person remembers is "the period I considered quitting, but stayed because of family." Narrative Compression converts Graph Truth into human-readable story arcs: cause → transition → outcome.

Phase 16 · 2026.06 · published
COMING SOON

Uncertainty Is Preserved, Not Eliminated

The system that pretends to certainty has stopped thinking. A position paper on epistemic humility in cognitive AI — and why confidence scores are not answers.

In preparation
// note   Full design documents and implementation notes are maintained in DESIGN_NOTES.md and XMN_CONSTITUTION.md. Contact xmn.ai.official@gmail.com for collaboration.

// STEP 8 · ORIGIN

Why this was built

XMN did not begin as a research project.
It began as a practical problem.

Started as a tarot app bug.
A voice engine separated from memory.
Memory needed to explain itself.
Explanation became a constitution.
That architecture is XMN.

2026.05
The Separation Problem
Repeated STT failures revealed a deeper issue:
voice processing and memory should never share the same responsibility.
2026.06.22
Birth of XMN
Voice Engine became independent.
Memory was no longer tied to audio.
2026.06.24
Memory as Architecture
The first Pure Memory Engine.
Graph Memory, Lifecycle, and Zero-Identity principles established.
2026.06.26
A Name for the Architecture
XMN — eXtensible Memory Node-layer.
The architecture received its identity.
2026.06.27
Thinking Beyond Memory
MPI. Autonomic Engine. Reinterpretation.
Memory evolved into a cognitive architecture.
2026.06.29
The Constitution
Truth. Evidence. Signals. Interpretation. Knowledge.

XMN became an Evidence-based Cognitive Architecture.
Next
Meaning Trajectory
Life Chapters.
Self-Evolving Interpretation.

// the architecture continues.

CONTACT

PROJECTXMN — eXtensible Memory Node-layer
DOMAINxmn-ai.com
EMAILxmn.ai.official@gmail.com
TIERLite · Pro · Enterprise
SDKpip install xmn  ·  coming soon
ORIGINVeiledNote → XMN (2026.06)

// What changed

It started as a tarot app.
A single bug led to a separation.
That separation revealed a memory engine.
That engine needed to explain itself.
That explanation became a constitution.

Uncertainty is preserved, not eliminated.