Reading Papers Without Drowning

Every data scientist I know has the same guilty browser tab: a reading list of papers thirty items deep, growing faster than it shrinks. For a long time mine did too. What finally helped wasn't reading more — it was reading with a system, and giving myself permission to stop early.

The three-pass rule

I steal this framing from a well-worn piece of advice: read a paper in up to three passes, and let each pass decide whether the next is worth it.

Pass one — five minutes. Title, abstract, figures, conclusion. Enough to answer: what problem, what claim, do I care? Most papers stop here, and that's the point.
Pass two — half an hour. Read for the idea, not the algebra. Follow the method at a block-diagram level, note the datasets and baselines, and mark the equations you'd need to revisit.
Pass three — a couple of hours. Only for papers you'll build on. Re-derive the key results, question every assumption, and try to imagine the experiment that would break the claim.

Give yourself an exit

The system works because it makes quitting a feature, not a failure. A paper that doesn't survive pass one was never going to be worth two hours.

Keep a second brain

Reading without writing evaporates. For every paper that makes it past the second pass, I keep three sentences in a note:

The one-line claim.
The trick that makes it work.
What I'd steal for my own work.

Six months later, those three sentences are worth more than the PDF.

The goal isn't to read every paper. It's to build a map good enough that you know which paper to read when the moment comes.

Where this connects

The habit feeds directly into the notes on this site — most of them started as a pass-three scribble that grew up. If you have a system that works better, I'd genuinely like to hear it.