Most papers I failed to learn from failed for the same reason: I read them linearly, once, and tried to hold everything at once.
Pass one: does the problem matter?
Title, abstract, figures, conclusion. Fifteen minutes. The only question is whether the problem is one I have. If not, stop — that is a complete and successful reading.
Pass two: what is the idea?
The introduction and the figure that carries the contribution. Ignore the related work, ignore the experimental setup. You are looking for the one sentence that could not have been written by a different paper.
The contribution is: <one sentence, in your own words>
If you cannot write it after pass two, you have not understood the idea yet, and reading onwards will not help.
Pass three: is it true?
Now the methods, the experiments, the ablations, and the limitations. This is where most papers are quietly wrong, and where the limitations section tells you whether the authors know it.
The question, written down first
Before pass one, write down what you expect. Not a guess at the answer — a question you actually have.
"Does their energy claim account for the ADC, or only the array?"
Then read to answer that specific question. You will get more out of ten papers read this way than a hundred read linearly, because you are reading against a hypothesis instead of just consuming text.
What to do with the ones that matter
Keep three things per paper: the problem, the idea, and the one experiment you would run to test it yourself. Everything else is detail you can re-derive.
- reading
- methodology
- writing