November 9, 2026

Evidence Matrix Guide: From Paper Collection to Synthesis

An evidence matrix is a working table that maps each paper you have collected to the specific claim, method, or gap it can support in your own writing. It is the bridge between a folder of PDFs and a synthesis you can de

Written byWisPaper TeamAI Research Workflow Team
Editorial illustration for Evidence Matrix Guide: From Paper Collection to Synthesis

An evidence matrix is a working table that maps each paper you have collected to the specific claim, method, or gap it can support in your own writing. It is the bridge between a folder of PDFs and a synthesis you can defend. If you are finishing a course paper, a thesis chapter, or a first manuscript, the matrix is what turns reading into an argument.

The practical goal is simple: every sentence of synthesis should trace back to a row you can point to. This guide diagnoses where you are starting, shows the decisions and artifacts each stage requires, and explains how to know the matrix is ready to write from.

What stage are you actually in?

Most people who feel stuck are not stuck on writing. They are stuck because collection, screening, and synthesis have blurred together. Use this quick diagnostic before you open another PDF.

Symptom you recognizeLikely stageFirst move
Hundreds of saved papers, no clear questionCollection without criteriaWrite eligibility criteria before adding more
A reading list but no notes you can compareScreening without extractionBuild the matrix columns first
Notes exist but every paper says something differentExtraction without a claim structureDefine the claims your chapter must support
You can summarize papers but not argueSynthesis not startedGroup rows by claim, not by author
A draft exists but citations feel shakyVerification pendingCheck references against real records

Cochrane's guidance on searching and selecting studies starts from eligibility criteria, not from a search box. That ordering matters: criteria decide what enters the matrix, and the matrix decides what your synthesis can claim.

Decide the question and eligibility criteria first

Before collecting, write down what a paper must do to earn a row. Useful criteria are concrete: population or setting, method type, time window, and the kind of evidence you will accept. Cochrane recommends planning searches from these criteria and combining free-text terms with subject headings, then documenting the process so it can be repeated.

NIH guidance on scientific rigor treats careful assessment of prior knowledge, strengths, weaknesses, and knowledge gaps as core research practice. Your eligibility criteria are where that assessment becomes operational: they force you to say which prior work counts and why.

A workable artifact here is a one-page protocol: the question, the criteria, the sources you will search, and the date you searched. Keep it next to the matrix. When a supervisor asks why a paper is missing, the protocol answers before you do.

Collect broadly, then screen in stages

Collection and screening are separate jobs, and mixing them is the most common cause of a bloated, unusable matrix.

Search with a documented strategy

Cochrane recommends using multiple sources, documenting the search, deduplicating records, and screening in stages. If you are starting a January project, run the search once with your criteria in hand, save the query and date, and resist the urge to keep topping up the list mid-synthesis. Rerunning searches before you finalize is a recognized step, so plan for one refresh rather than continuous drift.

For finding papers through a scholarly interface, WisPaper search is the place to run queries against academic sources; English readers can start from the WisPaper Scholar Search entry point. The point is not to collect everything — it is to collect against criteria you can state.

Screen in two passes

A title-and-abstract pass removes obvious mismatches. A full-text pass confirms that the paper actually reports what the abstract implied. Cochrane's staged screening model exists precisely because these are different judgments. Record why each excluded paper was excluded; that log becomes the audit trail behind your matrix.

One study can appear as several reports. Cochrane recommends linking multiple reports of a single study rather than counting them as independent evidence. In the matrix, give the study one row and note the linked reports, so your synthesis does not double-count the same finding.

Build the matrix: columns that earn their place

A matrix that tries to capture everything becomes a second reading list. Choose columns that map to the claims you need to make.

ColumnWhat it holdsWhy it earns its place
Study ID and citationStable identifier and full referenceLets you trace every claim back
Question or aimWhat the authors set out to doSeparates intent from result
Method and sampleDesign, setting, sizeSupports comparability judgments
Key findingThe result you may citeThe unit of synthesis
LimitationsStated and evident weaknessesFeeds your gap analysis
Claim supportedWhich of your claims this row servesTurns rows into an argument
Verification statusWhether the record was checkedFlags rows still to confirm

Keep the "claim supported" column narrow. If a row supports three different claims, split it or note the specific finding each claim rests on. This is where NIH's emphasis on strengths, weaknesses, and knowledge gaps becomes practical: the limitations column is your gap analysis in raw form.

Where does the matrix live, and how do you keep it usable?

A matrix is only useful if you can find, reopen, and question the papers behind each row. That means the table and the library need to stay in step.

Save or upload the papers you keep into My Library so each row has a stable home. When you need to interrogate the collected set — for example, to check what several papers say about one method — Library QA lets you ask questions based on papers uploaded to My Library or added from search results. Use it to test a row, not to replace reading it.

For a project that spans weeks, keep the protocol, matrix, and notes together in Research Projects, where you can create a project and return to it. A winter-break cleanup is a good moment to consolidate scattered folders into one project before the next term starts.

From rows to synthesis: group by claim, not by author

A literature review organized author-by-author reads as a list. A synthesis organized by claim reads as an argument. The matrix makes the second possible because each row already names the claim it supports.

Three synthesis moves

  1. Convergence. Several rows support the same claim. State the claim once and cite the rows that support it, noting where methods differ.
  2. Divergence. Rows conflict. Report the conflict and the methodological difference that may explain it, rather than averaging the findings.
  3. Gap. The limitations column shows what no row addresses. That is your contribution, stated as a gap the evidence leaves open.

PRISMA 2020 provides reporting templates, including flow-diagram templates, for reviews that need transparent reporting of search, selection, and synthesis. Even a course paper benefits from the discipline: it forces you to show how the rows were selected.

A supervisor-feedback example

Suppose your supervisor returns a chapter draft with "this reads as a summary, not an argument." The fix is rarely more reading. Sort the existing rows by the "claim supported" column. You will usually find that most rows cluster around two claims and a handful support none. Move the unsupported rows to a parking area, strengthen the two clusters, and write the gap from the limitations column. The revision is now a restructuring task, not a rereading task.

Verify references before you finalize

A matrix built from search results can still contain records that do not match a real publication. Before submission, check the references themselves.

Crossref exposes publisher-deposited scholarly metadata — titles, authors, venues, dates, identifiers, and post-publication updates — which is why checking against real records is feasible. TrueCite checks BibTeX against real academic databases to flag hallucinated references, which fits the final verification pass on your reference list. Run it after the matrix is stable, so you are checking the references you will actually cite.

Longer tasks: when the matrix needs a heavier workflow

Some projects outgrow a single table. A first manuscript with a broad scope, or a thesis chapter that must survey an unfamiliar area, may need a structured literature review before the matrix can be filled.

Scholar Agent supports longer research tasks in WisPaper's literature review, inspiration, and reproduction use cases. Use it when the collection step itself is the bottleneck, then bring the output back into your matrix and screen it against your criteria. If the next step is choosing a direction rather than summarizing one, Idea Discovery takes research topics, background, references, and experimental constraints to generate research ideas — a fit for a January project start where the question is still forming.

How do you know the matrix is ready?

Readiness is a set of checks, not a feeling. Run these before you draft.

CheckPass condition
Every synthesis claim has rowsNo claim rests on zero rows
Every row names a claimNo orphan rows in the main table
Multiple reports are linkedOne study, one row
Exclusions are loggedYou can explain any missing paper
References are verifiedEach citation matches a real record
The gap is statedThe limitations column yields a specific open question
The search is documentedQuery, sources, and date are recorded

If a check fails, fix that check rather than adding papers. Adding rows to a matrix with orphan entries usually deepens the problem.

Routes for different starting points

Readers arrive at this guide from different places. Pick the route that matches your situation.

  • Course paper due soon, reading list already set. Skip collection. Build the matrix columns, extract from the set you have, and verify references. Your constraint is time, not coverage.
  • Thesis chapter with supervisor feedback. Start from the feedback, sort rows by claim, and rewrite around clusters. Use the gap column to state your contribution.
  • Winter-break cleanup. Consolidate scattered PDFs into one library and one project, write the protocol, and screen what you already have. The goal is a clean starting state for the next term.
  • January project start. Write criteria first, run a documented search, screen in stages, and build the matrix as you read. This is the route where a structured literature workflow pays off most.

Your next concrete action

Open your current reading list and create the matrix with just three columns: citation, key finding, and claim supported. Fill ten rows from papers you have already read. If you cannot name the claim a row supports, that paper is not yet evidence for your argument — park it and move on. Ten filled rows will tell you more about your synthesis than another week of collecting.

FAQs

There is no fixed number. The right size is the number of rows your claims require, with each row traceable to a verified record. A matrix with a few well-extracted rows that map cleanly to your claims is more useful than a large table with orphan entries. Let the claim structure set the size.