Approach chart: reading a paper in the order you need

A research paper is not built as a continuous text. It is assembled from parts written for different readers: one for the reviewers, one for the people who would have to repeat the work, one for a database. Those parts are then arranged in a fixed order of headings that has nothing to do with the order in which they are useful to you. Reading such a document from the first word to the last treats it as prose, and an hour of honest effort ends in a highlighted file and no usable sentence about what was actually claimed.

Every section was written for somebody else

The abstract is written last and works as an advertisement. It has to survive a database search, so it compresses the claim and leaves out most of what would qualify it. It tells you what the paper wants to have found.

The introduction places the work in a conversation you are usually not part of, and many of its citations exist to show that the authors know the field. The results are the actual content: this is where something is asserted with evidence attached. The discussion is where the authors tell you what to conclude, and its final paragraph, the one about limitations, is often the most honest text in the document.

Nothing in that arrangement was designed for a reader meeting the topic for the first time.

The order of use, not the order of print

An approach chart is not read from the top of the page down. There is a fixed order determined by what the crew has to decide, and it runs against the layout: first the minima and the runway, because that settles whether the approach can be flown at all, then the procedure itself, the tracks and the altitudes, then the details, frequencies and lighting.

Papers can be taken the same way. Start with the abstract, which costs ninety seconds and gives you the claim. Go to the figures and tables next, captions included, because a well made figure carries the finding faster than the three pages of prose built around it. Then the discussion or the conclusion, where the claim appears in full sentences and gets qualified. Only after that, and only if you still need it, the introduction for context. The methods come last, and often not at all.

In fifteen minutes you have the claim, the evidence and the main weakness.

Twelve papers and one afternoon

The reading list is the real problem. Twelve papers, an afternoon, and the silent assumption that each of them has to be read. That assumption is what makes the list unreadable.

Run a triage pass first. Three minutes per paper: title, abstract, figures, last paragraph of the discussion. Then sort into three piles. Discard, which will be the largest. Skim, meaning you take one result and move on. Read, which should end up being two or three papers out of the twelve.

Write one line next to each discarded paper saying why it was discarded. Without it you will open the same file in three weeks and reach the same conclusion.

A paper that keeps appearing in the reference lists of the others goes into the read pile even when its own abstract looks unremarkable.

What you leave dark on the first pass

On a first pass through an unfamiliar paper you will not understand the notation, the statistical procedure or roughly a third of the terms. That is the normal condition and not a sign that you started too early.

The mistake is to repair it as you go. With a search window open beside the text, every unknown word costs four minutes and produces two more, and forty minutes of reading becomes three hours that end without the claim.

Use a rule instead. Mark the term and keep moving as long as you can still say what the authors are asserting. Stop and look something up only when the unknown thing blocks the claim itself, which happens more rarely than it feels. Afterwards, look at your marks. The terms that appeared three or four times are the vocabulary of the field and worth an hour. The ones that appeared once belong to that paper and can stay dark.

When the methods section becomes the point

There are situations in which the methods stop being optional. When you intend to build on the work. When the result surprises you, because a surprising result is more often an artefact of the method than a discovery. And when the number you want to quote depends on conditions the abstract does not mention.

Even then you are not reading for completeness. Four questions carry most of it: what exactly was measured, on whom or on what, how many, and what was compared against what.

This is usually where the claim narrows. The abstract says a method improves outcomes. The methods section says it was tested on one dataset, under conditions the authors chose, against a baseline nobody uses. Both statements sit in the same paper, and only one of them is quotable.

What is left when the file is closed

A paper that leaves nothing behind was not read, it was visited. Highlighting does not count. A marked passage is a promise to think about something later, and the later rarely arrives.

Write three or four lines while the file is still open. The claim in your own words, not in the authors’ phrasing, because their phrasing will let you write a sentence you cannot defend. What the evidence actually is, in one clause: an experiment on so many participants, a model on such and such data, an argument with no data. One limitation, ideally the one the authors admit to. And the reason this paper is in your collection, which is the line you will reach for when you write.

The test is simple. If you cannot produce the claim line without opening the file again, the reading did not take.

The papers you do have to read from front to back

All of this is triage, and triage is for the eighty papers you have to survive. It is the wrong treatment for the three or four that sit at the centre of your own work: the paper your thesis argues against, the method you are going to implement.

Those get read the way they were written, in order and more than once, including the appendix and the parts the triage pass is designed to skip, because in your own core topic the value has moved into the details. The equation you skipped is the equation you will have to modify.

Knowing which pile a paper belongs to is itself a judgement, and you will get it wrong sometimes. The only repair is to go back later and read properly something you once filed as skimmed.