A search turns up 60 papers. Downloading them all takes 2 GB; reading them all takes a month.
My old bad habit was "if I downloaded it, I have to read it." As a result, I read twenty papers before realizing only five were actually useful. The rest of the time went entirely into "reading a paper that wasn't worth reading."
Now I give myself three minutes, run three checks, and then decide: read closely, make a card only, or give up.
Check One: Are the Four Elements in the Abstract All There? (One Minute)
I look for only four things: research subject, method, main results, conclusion.
If two or more of the four are missing, the paper is most likely a short communication or a conference abstract with insufficient information, so it goes first into the "card only" pile.
If all four are there but the research subject is far from my topic, it also goes into the card pile.
In this minute I do only one thing: judge "whether it overlaps with my question."
Check Two: Can I Understand the Figures and Figure Captions? (One Minute)
I go straight to the results section and look only at the figures and captions—without reading the main text.
If I can understand what the figures are saying, the paper is clearly expressed and worth reading; if I can't understand the figures, it may be poorly written, or I may lack the prerequisite knowledge.
These two cases must be separated: if I lack prerequisite knowledge, I fill in the concepts first and then come back; if it is poorly written, I give up.
I read figure captions in English papers slowly, so I put the captions separately into MuchSci's paper translation and scan the Chinese-English comparison. MuchSci supports submitting multiple materials together, so I can go through the captions of several papers at once.
Check Three: Are There "Classics I Recognize" in the References? (One Minute)
This one is very practical: a serious study usually cites several foundational works in the references of this direction.
If I don't recognize a single one, the paper may be more applied or marginal—not unreadable, but lower priority.
Conversely, if it cites the few papers I am currently following, it is on the same line as me, so read it closely first.
The Three Destinations After Three Minutes
| Conclusion | Handling | Time |
| Read closely | Put on the "read thoroughly" list, read paragraph by paragraph | 1–2 hours |
| Card only | Write three lines: what it studies, the conclusion, what remains unsolved | 1 minute |
| Give up | Record the title, do not download | 10 seconds |
The key is that **"giving up" is also a decision**, and it should leave a trace. The benefit of leaving a trace is: the next time I search and find the same paper, I won't hesitate again.
My Actual Proportions
In a recent search of 60 papers: 6 read closely, 31 carded, 23 abandoned.
Of the 23 abandoned, 8 had "very similar titles but were not doing the same thing"—this type wastes the most time, because clickbait-style research is everywhere.
A Counterexample
Once I skipped these three minutes and read a 30-page review straight through. After finishing, I found it was about another system, and I used only one sentence.
If I had used those two hours to make cards, I could have gone through 30 papers.
A Little Reflection
Judging "whether it is worth reading" is more important than "reading fast." The three checks in three minutes are essentially asking three questions: does it overlap with my question, is it clearly expressed, and is it on the same line.
Of these three questions, I screen the first with MuchSci's paper search assistant (going through titles and abstracts first), and use its paper translation to look at figure captions for the second. It helps me make "looking" faster, but the decision "whether to read" is still mine to make—if I decide wrong, speed doesn't help.