Research Paper Reading

Overview

Generate structured reading notes from a research paper. The note structure follows how readers understand papers, not how authors write them:

Core Workflow

Three-pass reading, then note generation:

1. Skim pass (get the big picture): Title, abstract, all figures/tables, conclusion. Extract paper identity and one-sentence summary.
2. Comprehension pass (understand the story): Introduction for motivation, method for core idea, experiments for evidence. Extract problem, insight, method overview, key results.
3. Critical pass (evaluate and connect): Re-read method assumptions, experimental gaps, related work positioning. Extract strengths, weaknesses, connections, takeaways.

After three passes, produce the structured note using the template in references/note-template.md.

Note Structure (10 Sections)

Each section has a specific cognitive role. Do NOT reorder to match the paper.

1. Paper Identity

Metadata block. Title, authors, venue, year, link. No prose needed.

2. TL;DR (One-Sentence Summary)

Force-compress the entire paper into ONE sentence: "[This paper] does X by Y to achieve Z." This is the hardest and most valuable part. If you can't write this, you haven't understood the paper.

3. Problem & Motivation

• What specific problem does this paper address?
• Why is this problem important or timely?
• What is broken / missing / insufficient in prior approaches?

Do NOT copy the abstract. Restate in your own words as a reader.

4. Core Insight

The conceptual "aha" — the key idea that makes this paper work, stated BEFORE technical details.

• What is the one insight that distinguishes this paper from prior work?
• If you had to explain this paper to a colleague in 30 seconds, what would you say after "The key idea is..."?

This is NOT a method description. It's the intellectual contribution stripped of implementation. See references/extraction-guide.md for techniques.

5. Method

How the paper implements the core insight. Structure as:

1. Pipeline overview (one paragraph + describe the pipeline figure if available)
2. Key components (2-4 components, each with: what it does, why it's needed, how it works)
3. Training / optimization details (losses, objectives, important hyperparameters — only the non-trivial ones)

Skip standard/obvious components (e.g., "we use Adam optimizer with lr=1e-4" is usually noise).

6. Key Results

Cherry-pick the 2-4 most informative results. For each:

• What claim does this result support?
• What is the comparison baseline?
• What is the margin / significance?

Include ablation highlights if they reveal which design choices actually matter. Do NOT transcribe all tables — only the results that answer "does the core insight work?"

7. Strengths

What does this paper do well? Consider:

• Novelty of the idea
• Strength of experimental validation
• Clarity of presentation
• Practical utility / reproducibility

8. Weaknesses & Limitations

Be specific and constructive. Consider:

• Unstated assumptions
• Missing baselines or datasets
• Scalability concerns
• Failure cases shown or implied
• Claims not fully supported by evidence

See references/critical-reading.md for a structured evaluation framework.

9. Connections

Position this paper in the broader landscape — this section replaces "Related Work" from the reader's perspective:

• What prior work does this build on most directly?
• What concurrent or subsequent work addresses similar problems?
• How does this change your understanding of the field?

This section is meaningful only AFTER understanding the paper, which is why it comes late.

10. Takeaways & Open Questions

Personal reflection — the part that makes notes useful months later:

• Takeaways: Ideas to try, techniques to borrow, lessons learned
• Open questions: What remains unclear? What would you ask the authors?
• Follow-up reads: Papers referenced that you should read next

Execution Rules

1. Always perform the three-pass reading before writing notes. Do not start writing after the first pass.
2. The TL;DR is mandatory and must be exactly one sentence. Iterate until it is precise.
3. Core Insight must be stated in conceptual terms before any equations or algorithms.
4. Method section should use the paper's own figures as anchors when available.
5. Key Results must link back to claims — do not list numbers without stating what they prove.
6. Strengths and Weaknesses must be specific — "well-written" or "limited experiments" is not useful.
7. If the paper is behind a paywall or unreadable, report what you can extract and what is missing.
8. Output the notes as a single markdown file. Use the template in references/note-template.md.
9. Adapt depth to the paper's complexity: a 4-page workshop paper needs less detail than a 15-page journal paper.

Quality Checklist

Run this after generating notes:

• TL;DR is exactly one sentence and captures the core contribution
• Problem & Motivation is in reader's words, not copied from abstract
• Core Insight is conceptual, not procedural
• Method describes the pipeline before diving into components
• Key Results link to specific claims
• Strengths and Weaknesses are specific and actionable
• Connections position the paper relative to at least 2-3 other works
• Takeaways include at least one concrete idea or question
• No section is just a restatement of another section

Section Guides

Load only when needed:

• Note template: references/note-template.md
• Extraction guide (how to find core insights): references/extraction-guide.md
• Critical reading framework: references/critical-reading.md

Output Contract

When asked to generate reading notes, return:

1. The complete structured note following the 10-section template.
2. A confidence indicator for each section: [high] if well-supported by the paper, [medium] if partially inferred, [low] if the paper doesn't provide enough information.
3. A short list of suggested follow-up papers (extracted from references) if the user wants to dive deeper.