How to read a paper

There is far more research than anyone can read. The skill isn't reading fast — it's deciding what deserves a deep read, and getting the most from the ones that do. Here's a method that scales.

Three passes

Pass 1 — the five-minute scan

Read the title, abstract, section headings, and figures. Read the conclusion. Then answer:

• What are they claiming?
• What kind of contribution is it — a new method, a result, a critique, a survey?
• Is it relevant to what you care about right now?

Most papers should stop here. That's not failure — it's triage. A no after five minutes is a good outcome.

Pass 2 — the one-hour read

For the ones that pass, read the whole thing, but skip heavy proofs and derivations. Focus on:

• The method: what did they actually do?
• The evidence: what experiments or arguments back the claim?
• The figures and tables: this is where results live. Read them carefully; check that they show what the text says they show.

By the end you should be able to summarize the paper to a colleague — claim, method, evidence, and one limitation — from memory.

Pass 3 — the deep read

Only for papers you need to build on, reproduce, or review. Read as if you were re-implementing it. Re-derive the key steps. Note every assumption. Ask: if I ran this, what would break?

This pass is slow on purpose. A paper you've read this way, you own.

Read like a skeptic, not a cynic

• Check the baselines. A gain over a weak baseline is not a gain.
• Look for the ablation. Which part actually causes the effect?
• Mind the gap between claim and evidence. Did they measure what they say they measured, on the population they generalize to?
• Find the limitations section — and if there isn't one, be suspicious.

Skepticism is checking the work. Cynicism is dismissing it unread. Do the first.

Keep notes you'll actually reuse

For anything past pass 1, jot four lines: claim, method, evidence, weakness. That's enough to recall the paper months later and to write a fair critique if you want to respond to it here.

Reading well is upstream of writing well. The sharpest research on this site comes from people who read a lot of it first.