Reading a paper that matters, in 20 minutes

There are too many papers. There will keep being too many papers. You can read every abstract and still get nothing. Or you can read three papers a month deeply and actually know what you think.

The 20-minute pass

• 2 min — Title + abstract. What are they claiming?
• 3 min — Figures 1 and the results tables. Does the claim hold up if you only look at the data?
• 5 min — Methods section, scanning for: dataset size, baseline they compare against, the specific thing that's novel.
• 5 min — Limitations section (if it exists) and the discussion. What did they NOT test?
• 5 min — Write three sentences: what's the claim, what's the evidence, what would make you believe it more.

If you can do all five steps, you have a working read. If step 5 is hard, you didn't actually understand it — go back to steps 2 and 3.

What to skip

Most related work. Most citations. Pretty much all of the intro after the second paragraph. They're there for the reviewers, not for you.