How do I triage 200 PDFs before deciding which ones to read in full?

The mistake that makes a pile of PDFs unmanageable is treating reading as binary. Either you have read a paper properly or you have not, so every unread file carries an obligation, and two hundred obligations is a reason to open none of them.

Triage removes the obligation by making a decision the output. After five minutes you know what this paper is for, and that is a completed piece of work even if you never open it again. The pile stops growing in your head.

The five-minute pass has a fixed route. Title tells you the claim. Abstract tells you the design and the result. The last paragraph of the introduction tells you what the authors think they contributed, which is often clearer than the abstract. Figures tell you what was actually measured. The conclusion tells you how far they were willing to generalize. Skip everything else.

Adjust the route to your purpose. Carey and colleagues make picking your reading goal the first rule, and the goal changes what you look at: if you are entering a new field, weight the introduction's motivation and the conclusion's next steps, and if you are evaluating a technique, go to the methods and the reason the authors chose that approach (Carey et al., 2020, p. 2). Beginners get a stricter instruction, examine each panel of each figure separately rather than the figure as a whole.

Do the whole pile before reading anything in full. It is tempting to start reading the first interesting paper, and it wrecks the process, because you lose the comparative view that lets you see which five papers actually matter. Triage first, read second.

Expect the distribution to surprise you. In most piles, ten to fifteen percent deserve a full read, a further quarter contain one usable fact, and the rest are context you already have. Knowing that in advance makes the discard decisions much easier.

Those proportions have a measured analogue in systematic review screening. Across 25 reviews and 329,332 screening decisions, 18.07 percent of citations survived abstract screening and only 5.48 percent survived full text screening (Wang et al., 2020, p. 4). Your pile will not match those rates, since that evidence comes from clinical reviews with formal protocols rather than from an open literature search. The shape holds anyway, most of what you triage will never earn a full read.

What is the three-pass method for reading a paper?

First pass, five minutes on title, abstract, headings, figures, and conclusion, to decide whether to continue. Second pass, about an hour on the argument and evidence, skipping proofs and technical detail. Third pass, several hours reconstructing the work in enough detail to reproduce or challenge it. Most papers never leave the first pass.

The method comes from computer science, where it is widely taught, and it transfers to any field with a conventional paper structure. What makes it work is that each pass ends in a decision rather than in a feeling of incompletion.

The third pass is rare and should be. Reserve it for the handful of works your own contribution stands directly on, plus anything you intend to argue against, since arguing against a paper you have only skimmed is how vivas go badly.

There is no agreed order inside the first pass. In a survey of 88 researchers, 60 recommended selective reading and 28 recommended reading the sections in a specific order, but the recommended orders conflicted, with abstract first and figures first both common and no consensus on which is better (Hubbard and Dunbar, 2017, p. 8). Five respondents used the abstract or the discussion specifically to decide whether a paper was worth detailed reading, and 30 recommended reading critically, of whom 18 said to check whether the conclusions actually match the data (Hubbard and Dunbar, 2017, pp. 8-10). Pick an order, apply it to the whole pile, and save the critical check for the papers you promote.

How do I record a decision not to read something?

Write one line: the paper, the date, and why. "Wrong population", "superseded by the 2024 replication", "cited everywhere but no primary data". Without the reason you will reopen it in three months, and the reasons together become the inclusion criteria for your review.

Keep discards in the same place as everything else rather than deleting them. A discarded paper that turns out to matter later is easy to recover if the record exists, and impossible to find if you removed it from your library in a moment of tidiness.

Assume a share of your verdicts are wrong. Across 85 reviewers, the average abstract screening error rate, counting wrong inclusions and wrong exclusions together, was 10.76 percent, ranging from 5.76 percent to 21.11 percent depending on the clinical area and the type of question (Wang et al., 2020, p. 4). So classify the first twenty or thirty papers, check the borderline calls against your supervisor or a peer, and fix your criteria before you run the remaining hundred and seventy under rules you have not tested. The same authors put automated screening tools at a 30 to 70 percent workload reduction while warning that automation introduces its own errors and that human screening is an imperfect benchmark to measure it against (Wang et al., 2020, pp. 2, 6).

The accumulated reasons are worth more than they look. Read them together after a hundred papers and you will find you have been applying criteria you never wrote down, which is exactly the paragraph your methods section needs.