Read Your First Research Paper and Leave With a Usable Note

To read your first research paper, start with a question you want it to answer. Get an overview, follow one relevant claim into the results and methods, then write a note that separates the authors' claim, the evidence you checked, and what you still do not understand. You can finish a useful first reading with an unresolved question. You cannot finish it by quietly replacing that question with a guess.

Suppose a tutor assigns a paper before a discussion. Your immediate goal might be to explain what the authors tested and ask one specific question about the evidence. That requires more than an abstract, but less than reproducing every experiment. An assignment, journal club, or research project may require deeper reading; use its actual requirements as your finish condition.

Kognivu can help turn this learning goal and your available time into daily quests. Keep the PDF and reading notes in your own files. Plan your first paper-reading sessions in Kognivu, then check that the proposed tasks include the evidence you need to inspect. Kognivu is a planning tool here, not a research-paper verifier.

Choose the question and the depth

Before opening another tab, write why you are reading this paper. A useful question names something you can check:

What did the authors compare, what result supports their main claim, and under what conditions did that result hold?

For a methods assignment, you might instead need to explain an experimental setup. For a theory paper, you may need to trace an assumption through an argument. A survey asks a different question again: how does it organise the existing work? Do not force all three into the same checklist.

MIT's Communication Lab guidance ties reading order to the information you need. That is a useful starting principle: choose the depth for your task before treating every unfamiliar paragraph as equally urgent.

Pick one accessible paper suggested by your course or an instructor, preferably with a topic you already know a little about. Save its title, authors, year, stable link or DOI, and version. If you read a preprint, label it as one. Keep enough identity information to find the same document later.

Make an overview before digging into details

Read the abstract, the problem described in the introduction, the headings, and the conclusion. Look at the figures and their captions. The order can change with the paper and your purpose. At this stage, make a provisional outline rather than a verdict:

Paper identity:
My reason for reading:
Question the authors address:
Approach or comparison:
Main claimed result:
Place I need to inspect next:

S. Keshav's How to Read a Paper describes reading in passes, with an overview before closer attention to evidence and, when needed, a deeper reconstruction of the work. A first pass helps you decide what deserves further reading; it does not establish that the conclusion is correct.

If you cannot identify the research question, mark the sentence where you got lost. Look up the one background concept that prevents you from understanding it, or ask your tutor whether a more introductory paper would fit. Collecting twenty definitions without returning to the question can consume the whole session.

Trace one claim to its evidence

Choose the claim most relevant to your purpose. Find the result, table, figure, argument, or cited source that supports it. Record a locator such as “Figure 2, caption, and Results section 3.1.” Page numbers can differ between versions, so a section or figure number helps.

For a figure, check the axis labels, units, legend, caption, and conditions. Then read enough of the methods to understand what produced that display. Ask what was measured, what was compared, and what the authors held constant. If a symbol or uncertainty measure is unexplained, record that gap; the mere presence of error bars does not settle the interpretation.

Here is an invented reading exercise, not a real study. Imagine a computing paper compares two file-processing methods:

Detail in the fictional paper	What your note should preserve
Figure 2 shows median runtime: method A, 12 seconds; method B, 9 seconds	B has a lower median runtime in this reported comparison
The caption specifies 100 MB files on one stated machine	The observation belongs to that workload and setup
The methods say each method ran ten times	Ten runs are reported; this alone does not establish how much the results vary
You cannot find an explanation of whether the file cache was reset	Cache handling remains an unresolved methods question

The difference is three seconds. Relative to A's twelve seconds, that is 3/12, or 25% lower median runtime. Write the denominator and the measure. “B is 25% better” hides what improved and under which conditions.

Do not extend the result to every file size or machine. Do not infer statistical significance from the two medians. Your first reading can explain the reported comparison while leaving its variability and cache handling open.

For a proof or conceptual argument, replace this numerical check with a chain of statements: assumption, intermediate step, conclusion. Mark the first step you cannot justify. You need a question about that step, not a pretend experimental table.

Write a note another reader can check

A useful note keeps interpretation attached to the source. For the fictional example, it could read:

Purpose: Explain the reported runtime comparison before the discussion.
Authors' claim: Method B reduces runtime for the tested workload.
Evidence checked: Figure 2 and its caption; relevant methods paragraph.
Observation: Median runtime is 9 s for B and 12 s for A.
Calculation: (12 - 9) / 12 = 25% lower median runtime relative to A.
Scope: 100 MB files, stated machine, ten runs per method.
Unresolved: I could not establish cache handling or interpret run variability.
Question: Where is cache reset described, and could it affect this comparison?
Next reading: Recheck setup details and any referenced supplement.

Now close the paper and explain the question, approach, and one result in your own words. Reopen it to check what you said. Repair any part of the note that adds a conclusion the paper did not support or loses an important condition.

If you use an AI summary as an orientation aid, treat it as a separate proposal. Locate each claim you plan to keep in the actual paper. A confident summary or a plausible page reference is not a substitute for finding the passage. Your note should make visible which evidence you checked yourself.

For a discussion, share your specific unresolved question. “I don't understand the paper” gives the other person little to work with. “I found ten reported runs, but cannot tell how the cache was handled” identifies the next useful explanation.

Split the remaining reading into finishable sessions

If you have four half-hour blocks this week, try this starting budget: one for the overview, one for the claim and figure, one for the relevant methods and background, and one for the note and explanation check. That is two planned hours, not a promise that every paper will fit. A difficult proof, unfamiliar method, or missing prerequisite can require more time.

After each block, save the next precise action: “Read the setup paragraph to find the cache rule” is easier to resume than “understand the paper.” If a session slips, preserve your note and locator. Reduce an optional second claim or move the finish date before dropping the evidence check.

The learning curriculum guide helps place background study into a broader goal. If this reading belongs to a course you keep abandoning, the course-completion guide addresses returning to the unfinished work. Neither replaces the paper-specific question.

You are ready for the first discussion when you can identify the document, state the question and approach, trace one relevant claim to its evidence, explain the limits, and ask a specific remaining question. Full reproduction or expert critique is a later task when your purpose requires it.

Build my paper-reading goal in Kognivu with your current background, discussion date, remaining checks, and available sessions. Review the generated path and first quest against the actual paper; keep the evidence and unresolved questions in your reading note.