Great-notes Core Protocol

Great-notes is Mortis's learning knowledge base. Each domain is organized into three layers:

1. raw/: input material, including human-added files and tool-converted files.
2. progress/: shared human-AI interaction workspace.
3. wiki/: AI-retained durable knowledge state.

The core objects are:

• progress/learning_log.md: complete learning log, editable by human and AI.
• wiki/concepts/: Knowledge Nodes, maintained by AI.

Everything else is a later abstraction or generated view.

---

Domain Layout

{domain}/
├── raw/
│   ├── papers/       human-added papers and articles
│   ├── textbooks/    books, textbook sources, and converted textbook text
│   └── clips/        notes, excerpts, screenshots, fragments
├── progress/
│   ├── README.md
│   └── learning_log.md
├── wiki/
│   ├── README.md
│   ├── concepts/     Knowledge Nodes
│   ├── archive/      archived old records
│   ├── index.md      future generated view placeholder
│   ├── tree.md       future generated view placeholder
│   └── progress.md   future generated view placeholder

Tool-converted material is input material. Put it under raw/, usually raw/textbooks/, unless Mortis explicitly defines a separate staging process.

---

Layer Responsibilities

raw/

Input layer.

• Humans place materials here.
• Tools may also place converted materials here when Mortis treats them as readable inputs.
• Humans may organize by papers/, textbooks/, and clips/.
• AI may read raw materials.
• AI must not rewrite raw materials.
• Source type is metadata only; it must not determine the learning workflow.

progress/

Shared human-AI workbench.

• Humans may write here.
• AI may write here.
• learning_log.md is the complete chronological learning record.
• Use this layer for session notes, direct interaction records, corrections, user intent, and full learning history.
• Do not summarize away history unless Mortis explicitly requests an archive.

wiki/

AI-retained durable knowledge state.

• AI maintains this layer.
• Humans normally do not need to edit it.
• wiki/concepts/ stores Knowledge Nodes.
• AI distills durable state from progress/learning_log.md and relevant raw materials into Knowledge Nodes.
• index.md, tree.md, and progress.md are placeholders for future derived views until Mortis defines them.

---

Primary Object 1: Learning Log

Path:

{domain}/progress/learning_log.md

This is the complete interaction and learning history. It is primary storage. Append events; do not overwrite old events.

Recommended event structure:

## Event YYYY-MM-DD-001
- Time:
- Trigger: user_requested / task_started / task_completed / review / import / correction
- Input:
- User intent:
- Main nodes:
- Secondary nodes:
- What happened:
- Evidence created:
- Mastery changes:
- Relation changes:
- Prerequisite debt discovered:
- Open questions:
- Next actions:
- Source references:
- Momentum references:

The log should preserve:

• what material was studied
• what Mortis asked, said, rejected, or corrected
• what the AI inferred
• what evidence appeared
• what changed in mastery or relations
• where Mortis jumped ahead
• what uncertainty remains
• what should happen next

---

Primary Object 2: Knowledge Node

Path:

{domain}/wiki/concepts/{node}.md

A Knowledge Node is a learnable unit: concept, theorem, method, skill, exercise type, misconception, open question, or recurring difficulty.

Required structure:

---
title: "Node Title"
type: knowledge_node
domain: Domain
created: YYYY-MM-DD
updated: YYYY-MM-DD
status: active
slug: /domain/path/node
sources:
  - "raw/..."
---

# Node Title

## Node Identity
- Kind:
- One-line meaning:
- Current role in learning:

## Position State
- Path hypothesis:
- Position confidence: high / medium / low / unresolved
- Position status: anchored / floating / skipped_ahead / prerequisite_gap
- Notes:

## Mastery State
- Level: unlearned / seen / confused / practicing / usable / fluent
- Confidence: high / medium / low
- Last evidence:
- Last updated:

## Evidence Ledger
- YYYY-MM-DD | event-id | evidence-type: summary

## Relations
### Prerequisites
- Node:
  relation confidence:
  why:

### Depends On Me
- Node:
  relation confidence:
  why:

### Related / Contrast / Analogy / Application
- Node:
  relation type:
  relation confidence:
  why:

## Prerequisite Debt
- [ ] Node:
  why it matters:
  discovered at event-id:

## Open Questions
- [ ] Question:
  discovered at event-id:

## Next Local Actions
- [ ] Action:
  reason:

Uncertainty rules:

• A node may be floating.
• A node may be skipped_ahead.
• A relation may have low confidence.
• Hiding uncertainty is worse than leaving a field unresolved.

---

Session Protocol

At session start:

1. Read this protocol.
2. Read domain MDclaude.md if present.
3. Read {domain}/progress/learning_log.md.
4. Read relevant Knowledge Nodes from {domain}/wiki/concepts/.
5. Ask or infer the learning intent.

During the session:

1. Keep the main learning thread visible.
2. Append meaningful events to progress/learning_log.md.
3. Create or update Knowledge Nodes for durable knowledge state.
4. Record mastery evidence when it appears.
5. Record relation changes with reasons.
6. Record jump-ahead and prerequisite debt immediately.
7. Keep tangents separate from the main thread.

At session close:

[ ] learning_log.md has a new event
[ ] Main Knowledge Nodes were created or updated
[ ] Mastery changes have evidence
[ ] Relation changes have reasons
[ ] Jump-ahead / prerequisite debt recorded if present
[ ] Open questions recorded if present
[ ] Next local actions recorded

The session is not complete unless the log can reconstruct what happened and the nodes can answer current state questions.

---

Derived Views

Do not maintain derived views unless Mortis asks.

Current placeholder views:

• wiki/index.md
• wiki/tree.md
• wiki/progress.md
• future graphs and dashboards

These must not be treated as source of truth. Later, Mortis may define how to generate them from:

progress/learning_log.md
wiki/concepts/

---

Momentum Integration

Momentum should read:

{domain}/progress/learning_log.md
{domain}/wiki/concepts/

Useful future queries:

getRecentLearningEvents(domain)
getNodeMastery(node)
getNodeRelations(node)
getPrerequisiteDebt(node or domain)
getCandidateNextActions(domain)

Momentum should not depend on wiki/index.md, wiki/tree.md, or wiki/progress.md until their rules are explicitly defined.

---

Style

• Preserve complete learning history in progress/learning_log.md.
• Preserve durable state in Knowledge Nodes.
• Prefer explicit uncertainty over invented structure.
• Use source type as metadata, not workflow control.
• Use Chinese, English, or both as clarity requires.

Final maxim:

raw is what Mortis gives the system, including tool-converted input.
progress is where Mortis and AI work together.
wiki is what AI keeps after distillation.