Reading research papers is a learned skill, not a natural one. Most people who try to read papers the same way they read textbooks find it slow, frustrating, and unproductive. The key insight is that papers are not meant to be read linearly. They have a structure that supports a different kind of reading, one that lets you extract value efficiently. This lesson covers the three-pass reading method, how to find a paper's key contribution, the workflow for reproducing results, and how to contribute back to the open source ecosystem.

Why Reading Papers Is Hard

Papers have several properties that make them hard to read:

1. Dense: Papers compress significant work into a few pages. Every sentence is doing a lot of work.
2. Self-referential: Papers often reference other papers without explaining them. To fully understand one paper, you may need to understand several others.
3. Specialized vocabulary: Each subfield has its own terminology. Without the vocabulary, papers are nearly opaque.
4. Variable quality: Some papers are written clearly, others are not. Reading skill cannot fix a poorly written paper.
5. Hidden context: The paper is written for an audience that knows what was happening in the field at the time. Reading later, you miss the context that informed the choices.
6. Confused goals: Are you reading to understand the contribution, to evaluate the work, to find techniques to use, or to find related work? Each goal calls for a different reading approach.

The three-pass method addresses these by giving you a structured approach that can be tailored to your purpose.

The Three-Pass Method

The three-pass method, popularized by Keshav's short article "How to Read a Paper", divides reading into three passes of increasing depth.

Pass 1: Get the Big Picture (5-10 minutes)

The first pass aims to answer: what is this paper about, what does it claim, and is it worth reading more carefully?

Read in this order:

1. Title, abstract, introduction
2. Section and sub-section headings
3. Conclusion
4. Glance at references

After pass 1, you should be able to answer:

1. What category of paper is this (theoretical, experimental, review, survey)?
2. What is the main contribution?
3. Is the paper well-presented?
4. Is it relevant to your interests?

The first pass is a triage step. Most papers you start reading should not get past it, they are not relevant enough to your goals to justify further time. Papers that survive the first pass deserve more attention.

Time budget: 5-10 minutes per paper. If you cannot make sense of the paper's claim in this time, the paper is poorly written or you lack background, either way, the next passes will be harder.

Pass 2: Grasp the Content (1 hour)

The second pass aims to understand the paper's content without dwelling on details. Read more carefully, examine figures, but skip detailed proofs and technical sections that you cannot follow.

In pass 2:

1. Read carefully but ignore details like proofs
2. Pay attention to figures, diagrams, and tables, they often convey the essence
3. Note the references for related work that might be useful
4. Mark sections you do not fully understand for the third pass

After pass 2, you should be able to:

1. Summarize the main thrust of the paper to someone else
2. Articulate the supporting evidence
3. Identify what you do not understand

Time budget: about an hour. This is enough to read a 10-page paper carefully but not enough to master every detail.

The second pass is where most of your "reading" time goes for papers worth reading. You skim less interesting parts, focus on the contributions, and absorb the figures. After this pass, you have a working understanding of the paper.

Pass 3: Master the Details (4-5 hours)

The third pass is for papers you really need to understand or reproduce. It involves working through the paper as if you were re-doing the work. You should virtually re-implement the paper in your head, making the same assumptions and choices the authors did.

In pass 3:

1. Read every word carefully
2. Verify the proofs and technical claims
3. Identify and challenge every assumption
4. Think about how you would do this differently
5. Make notes about what is unclear, missing, or potentially wrong

After pass 3, you should be able to:

1. Reproduce the work conceptually without referring to the paper
2. Identify the paper's strong and weak points
3. Propose follow-up research directions
4. Spot implicit assumptions and ungeneralized claims

Time budget: 4-5 hours, sometimes more. This is a major investment and is reserved for papers that really matter to your work.

When to Use Each Pass

The three passes serve different purposes:

1. Pass 1 is for triage. You are deciding whether the paper is worth more time.
2. Pass 2 is for general literature awareness. You want to know what ideas are out there without mastering each one.
3. Pass 3 is for the few papers you need to deeply understand: papers in your specific area, papers you want to build on, papers you need to replicate.

Most researchers do many pass 1s, fewer pass 2s, and very few pass 3s. The skill is in knowing which papers to invest more time in.

Making Pass 1 Faster

Pass 1 is the most common pass and the one that benefits most from practice. Some tips for being efficient:

1. Read the abstract carefully: A good abstract tells you the contribution. A bad abstract requires you to read the introduction.
2. Look at the figures first after the abstract: Often the main result is in a figure. The figure summarizes what the paper is claiming.
3. Read the conclusion: It usually restates the contribution and any limitations.
4. Identify the "main result" sentence: Most papers have a sentence that explicitly states the contribution. Find it.
5. Decide quickly: If the paper is not interesting, stop. If it might be, continue. Do not feel obligated to finish reading anything.

The goal of pass 1 is to filter out 80%+ of papers efficiently so you can focus your time on the remaining 20% that matter.

Reading for Different Goals

The three-pass method is a default. Specific goals call for variations:

1. Reading to find techniques: Skim for the methods section. Look at the experiments to see when the technique helps. You may not need to understand the theory.
2. Reading to write a related work section: Pass 1 is usually enough. You need to know the contribution and how it relates to your work, not the details.
3. Reading to reproduce: Pass 3 with attention to implementation details. The paper might not have everything you need, you may need to read the released code as well.
4. Reading to evaluate as a reviewer: A careful pass 2 with attention to claims, evidence, and logical gaps. You need to judge whether the paper is correct and contributes meaningfully.
5. Reading to learn the field: A series of pass 1s and a few pass 2s on the most influential papers. Build a map of the field, then dive deeper as needed.

Different goals are well-served by different reading strategies. Match the strategy to the goal.