| name | deep-dive |
| description | Integrated domain exploration combining vocabulary learning with research paper genealogy. Use when user says "deep dive [domain]", "explore [domain] deeply", "domain exploration", or wants to learn concepts with their research origins. |
| version | 0.1.0 |
Deep Dive: Concept + Paper Integration
Learn domain concepts while tracing their research origins. Combines domain-vocab (conceptual vocabulary) with paper-flow (citation genealogy) for contextualized understanding.
When to Use
- "Deep dive into [domain]"
- "Explore [domain] deeply"
- "Domain exploration"
- "I want to learn [domain] concepts and their papers together"
- Learning a new field with academic depth
When NOT to Use
- Quick concept lookup → use
domain-vocab
- Single concept history → use
trace
- Latest research only → use
frontier
- Citation tree only → use
paper-flow
Core Value
Learning concepts alone tells you "what". Learning with paper genealogy tells you "why, where, and how it evolved".
Workflow
Phase 1: Domain Entry (domain-vocab)
Objective: Extract core concepts with authentic expert tokens
Actions:
-
Invoke domain-vocab Phase 0-2:
- Token priming (key figures, literature sampling)
- Domain identification (scope, depth level)
- Core concept extraction (20-30 terms)
-
Identify anchor concepts (5-8):
- High-impact terms that shaped the field
- Concepts with clear research origins
- Terms that connect to seminal papers
Output:
domain: "[identified domain]"
depth_level: L2
concepts:
- name: "[concept]"
difficulty: entry|intermediate|advanced
is_anchor: true|false
potential_papers: ["keyword hints for paper search"]
User Checkpoint:
"I'll explore the research lineage for these anchor concepts: [anchors]. Proceed?"
Phase 2: Research Genealogy (paper-flow)
Objective: Build citation trees for anchor concepts
Actions:
-
For each anchor concept:
- Search papers using concept name + domain context
- Identify root paper (seminal work that introduced/defined the concept)
- Build citation tree (parents + children)
- Extract key papers (high-citation, influential)
-
Cross-link papers:
- Find papers that connect multiple anchor concepts
- Identify "bridge papers" that link sub-domains
Output per anchor:
anchor: "[concept name]"
root_paper:
title: "[paper title]"
authors: "[authors]"
year: YYYY
key_contribution: "[one sentence]"
citation_tree:
parents: [papers this cites]
children: [papers citing this]
key_descendants: [most influential follow-ups]
bridge_connections:
- connects_to: "[other anchor]"
via_paper: "[paper title]"
Phase 3: Integrated Output
Objective: Present unified view of concepts + papers
Output Formats:
Format A: Enriched Concept Cards
# [Concept Name]
**Difficulty:** Intermediate
**Domain:** [domain]
## Definition
[concept definition from domain-vocab]
## Research Origin
- **Seminal Paper:** [title] ([year])
- **Key Authors:** [authors]
- **Core Contribution:** [one sentence]
## Evolution
- **Preceded by:** [parent concepts/papers]
- **Led to:** [descendant concepts/papers]
## Practical Context
[from domain-vocab Phase 4]
## Related Papers
- [paper 1]
- [paper 2]
Format B: Obsidian Vault Structure
{domain}-deep-dive/
├── concepts/
│ ├── {concept-1}.md
│ ├── {concept-2}.md
│ └── ...
├── papers/
│ ├── {paper-id-1}.md
│ ├── {paper-id-2}.md
│ └── ...
├── {domain}-concepts.canvas # Concept relationship map
├── {domain}-papers.canvas # Citation tree visualization
└── {domain}-integrated.canvas # Combined view
Format C: Summary Report
# [Domain] Deep Dive Summary
## Core Concepts (by difficulty)
### Entry Level
- [concepts...]
### Intermediate
- [concepts...]
### Advanced
- [concepts...]
## Research Landscape
### Foundational Papers
| Concept | Seminal Paper | Year | Impact |
|---------|---------------|------|--------|
| ... | ... | ... | ... |
### Evolution Timeline
[Mermaid timeline diagram]
### Key Researchers
- [researcher]: [contributions]
## Learning Path Recommendation
1. Start with: [entry concepts]
2. Read: [foundational paper]
3. Progress to: [intermediate concepts]
4. Deep dive: [advanced + recent papers]
Decision Rules
Anchor Concept Selection
| Criterion | Weight | Example |
|---|
| Has clear seminal paper | High | "Attention" → Bahdanau 2014 |
| High interconnectedness | High | Appears in many concept relationships |
| User interest signal | High | User asked about this specifically |
| Intermediate+ difficulty | Medium | Entry-level often too broad |
| Recent research activity | Medium | Active area = more papers to explore |
Depth Calibration
| User Signal | Action |
|---|
| "quick", "overview" | L1-L2, 3 anchors, shallow trees |
| Default | L2, 5 anchors, 2-level trees |
| "deep", "thorough" | L2-L3, 8 anchors, 3-level trees |
| "researcher level" | L3-L4, all concepts as anchors |
Error Handling
| Situation | Recovery |
|---|
| No papers found for concept | Use broader search terms, or mark as "practice-origin" (not academic) |
| Concept too broad (1000+ papers) | Ask user to narrow scope |
| API rate limits | Queue requests, notify user of delay |
| Disconnected concepts | Note as "emerging" or "cross-domain import" |
Integration Points
With domain-vocab
- Uses: Phase 0-4 (token priming through practical context)
- Extends: Adds paper provenance to each concept
With paper-flow
- Uses: Phase 1-3 (discovery, network extraction, tree analysis)
- Focuses: Only on anchor concepts (not exhaustive field mapping)
Example Session
Input: "Deep dive into Reinforcement Learning"
Phase 1 Output (domain-vocab):
Domain: Reinforcement Learning
Depth: L2 (Practitioner)
Concepts (25):
- Entry: reward, state, action, policy, value function
- Intermediate: Q-learning, policy gradient, actor-critic, exploration-exploitation
- Advanced: PPO, SAC, model-based RL, offline RL
Anchors (5):
1. Q-learning (tabular foundation)
2. Policy Gradient (direct optimization)
3. Actor-Critic (hybrid approach)
4. Deep Q-Network (deep RL era)
5. PPO (practical standard)
Phase 2 Output (paper-flow):
Anchor: Deep Q-Network
Root Paper:
- "Playing Atari with Deep RL" - Mnih et al. 2013
- DeepMind, introduced DQN
Citation Tree:
- Parents: Q-learning (Watkins 1989), CNN for vision
- Children: Double DQN, Dueling DQN, Rainbow
- Bridge: Connects to "Actor-Critic" via A3C (2016)
Phase 3 Output:
- 25 enriched concept cards (5 with full paper genealogy)
rl-integrated.canvas showing concept-paper connections- Summary report with learning path
Future Extensions
/1d1s:trace [concept]
Trace a single concept's paper lineage in detail.
/1d1s:frontier [domain]
Focus on L4 (cutting-edge) with latest papers only.
Notes
- Phase 1 runs first, Phase 2 depends on anchor selection
- User can adjust anchors before Phase 2 starts
- Output format can be changed mid-session
- Progress is resumable (concepts extracted → papers pending)
