How to Build a Shared Evidence Folder for a Research Team

A shared evidence folder is the team-level record of what you searched, what you kept, why you kept it, and how each item bears on a defined question. It is not a pile of PDFs. It is a small, governed system: a question frame, a search log, a deduplicated record set, a screening trail, a linked-study map, and a status layer that tells every collaborator what is settled and what is still open.

This guide gives clinicians and medical researchers a complete sequence for building that folder, with checkpoints, a worked example, a reusable checklist, and stop rules. The running example is a focused population-and-outcome question: in adults with type 2 diabetes, what does the evidence say about cardiovascular outcomes associated with a specific drug class? The folder is built to support an evidence brief, not to issue individual treatment advice.

Most teams arrive at this task in one of three states. Naming yours determines how much structure you need.

• Scattered state. Papers live in personal download folders and email threads. There is no shared question statement, and no one can say which records have already been screened. You need the full sequence below.
• Partially organized state. A reference manager collection exists, but tags are inconsistent and the search history is undocumented. You need to retrofit a search log and a screening trail onto what exists.
• Rebuild state. A folder exists but has drifted: duplicates, multiple reports of the same study filed as separate studies, and no record of why items were excluded. You need to re-establish the question frame and re-run deduplication and linking.

A quick diagnostic: ask two team members to state the eligibility criteria from memory. If their answers differ, the folder has no usable frame yet, and everything downstream will inherit that ambiguity.

Cochrane guidance is explicit that searches should be planned from eligibility criteria. That means the folder's first artifact is not a paper but a one-page frame.

Write down, in plain language:

1. Population. Who is in scope, and who is out. In the example: adults with type 2 diabetes; exclude studies restricted to type 1 diabetes or to pediatric populations.
2. Exposure or intervention. The drug class, with enough specificity that a screener can apply it consistently.
3. Comparator. Placebo, usual care, or an active comparator — decide which are acceptable.
4. Outcomes. Which cardiovascular outcomes count, and over what follow-up.
5. Study designs. Which designs you will include, and which you will log but set aside.
6. Date and language limits. State them explicitly rather than letting them emerge by accident.

Checkpoint: every team member can restate the frame in two sentences, and the frame is stored as the first file in the folder. If the frame changes later, date the change and note what it affects.

Cochrane recommends using multiple sources, combining free-text terms with subject headings, and documenting the process. PRISMA 2020 provides reporting structures, including flow-diagram templates, for transparent reporting of search, selection, and synthesis methods.

Your search log should capture, for each search:

• the database or source searched;
• the full strategy, including free-text terms and subject headings;
• any limits applied;
• the date run and who ran it;
• the number of records returned;
• the number carried forward after deduplication.

This log is what lets a colleague reproduce your search or update it later. Cochrane also recommends rerunning searches before publication, so leave the log open rather than closing it after the first pass.

When you need to find candidate papers through a scholarly search interface, WisPaper search is one place to run and capture those queries; record the strategy in your log regardless of which interface you use, so the trail stays portable.

Cochrane describes deduplicating records and screening in stages. In practice, the folder should have distinct zones for each stage so that nothing is silently skipped.

• Raw records. Everything returned, untouched.
• Deduplicated set. One record per item, with duplicates merged and the merge noted.
• Title and abstract screened. Each record marked include, exclude, or uncertain, with a one-line reason.
• Full text assessed. Each included record checked against the frame, with reasons for any exclusion at this stage.

Two screeners at the full-text stage, with disagreements logged and resolved, is the pattern most teams will want for a brief that others will rely on. Record the resolution, not just the outcome.

Checkpoint: the count of records at each stage reconciles. If your raw count minus duplicates does not equal your screened count, you have a gap to close before moving on.

Cochrane recommends linking multiple reports of one study. This is the step most shared folders get wrong, and it distorts everything downstream: a trial reported in a primary paper, a secondary analysis, and a conference abstract can look like three studies if filed separately.

Create a study-level record that lists every linked report beneath it. The study, not the paper, becomes the unit you count and appraise. Your folder should make it obvious, at a glance, how many distinct studies you have versus how many documents.

Checkpoint: no study-level record has orphan reports, and no report appears under two studies without a note explaining why.

These four judgments are related but distinct, and collapsing them is the most common source of overstatement in an evidence brief.

• Evidence level describes the strength of the design and the body of evidence for the question — for example, whether the synthesis rests on randomized trials or on observational data.
• Applicability asks whether the studied population, setting, and care context resemble the one your team cares about.
• Indirectness flags mismatches between the evidence and your question: a different comparator, a surrogate rather than the outcome you specified, or a follow-up too short to capture the outcome.
• Limitations are the weaknesses within a study or body of evidence — risk of bias, imprecision, inconsistency.

Keep these in separate columns in your extraction sheet. A study can be high-level evidence and still be indirect for your question; a study can be directly applicable and still carry serious limitations. Writing them in separate columns forces the team to argue about each one on its own terms.

For the diabetes example, a trial in a broadly similar adult population with a composite cardiovascular outcome may be applicable, while a trial using a surrogate metabolic endpoint is indirect for the cardiovascular question regardless of its design strength.

A shared folder works best when each tool does one job and hands off cleanly.

• Finding papers. Run and log searches, then carry records into your deduplicated set. A scholarly search entry point such as WisPaper Scholar Search supports the discovery step.
• Saving and organizing. When you are ready to hold the kept set in one place, My Library lets you save or upload papers and organize them. Mirror the folder's zones there so the structure survives.
• Asking questions of the kept set. Once papers are in the library, Library QA lets you ask questions based on papers uploaded to My Library or added from search results. Use it to interrogate the set you have already curated, not to replace screening.
• Checking citation metadata. Before circulating a reference list, TrueCite checks BibTeX against real academic databases to flag hallucinated references — a useful gate on the bibliography artifact.
• Larger synthesis tasks. For a longer literature review or reproduction-oriented task, Scholar Agent fits the literature review, inspiration, or reproduction use cases in that area.
• Project-level coordination. If the folder belongs to a broader effort with milestones, Research Projects is where a new project can be created to hold that work.

Reference managers offer a complementary structure worth borrowing: collections can place one item in multiple project groupings without duplicating it, tags can describe topic, method, status, or workflow, and saved searches update automatically. A duplicate-items view supports review and merging. Whatever tool you use, the principle holds — one item, many views, no copies.

Here is how the sequence looks end to end for the diabetes question.

1. Frame. Adults with type 2 diabetes; the named drug class; comparators limited to placebo or usual care; outcomes limited to major cardiovascular events; randomized trials and large observational studies; no language limit; searches from a stated start date.
2. Search log. Three sources searched, each with free-text terms plus subject headings, dates and counts recorded.
3. Deduplication. Raw records merged to a single deduplicated set; the merge count is logged.
4. Screening. Title and abstract pass marks each record include, exclude, or uncertain with a reason; two reviewers assess full texts and log disagreements.
5. Linking. A primary trial report, its secondary cardiovascular analysis, and a conference abstract are linked under one study record.
6. Extraction. Each study gets separate columns for evidence level, applicability, indirectness, and limitations.
7. Synthesis. The brief states what the body of evidence shows for the specified outcome, with indirectness and limitations visible beside each claim.
8. Pre-publication rerun. Searches are rerun and any new records are screened before the brief circulates.

Checkpoint at the end: a colleague who was not on the team can open the folder, read the frame, follow the search log, reconcile the counts, see the linked studies, and understand why each excluded item was excluded — without asking anyone.

Use these to decide when to pause rather than push forward.

• Stop if the frame is contested. If team members disagree on population or outcome definitions, resolve that before screening. Screening against a moving frame wastes the trail.
• Stop if counts do not reconcile. A mismatch between stages means records were lost or double-counted. Find them before synthesis.
• Escalate when a study's applicability is disputed. Send it to a second reviewer with the specific mismatch named, rather than averaging opinions.
• Escalate when a linked-study decision is unclear. Whether two reports describe one study changes your study count; get a second read.
• Stop before circulating if the bibliography has not been checked. Run the citation check as a gate, not an afterthought.
• Rerun searches before the brief is finalized. New records found at this stage still need screening.

The folder is ready for a team discussion or evidence brief when all of the following hold:

• The question frame is dated, stored first, and restated consistently by the team.
• The search log is complete enough to reproduce, and the pre-publication rerun is recorded.
• Stage counts reconcile from raw records to full-text assessment.
• Every exclusion carries a reason, and full-text disagreements are logged with resolutions.
• Multiple reports of one study are linked under study-level records.
• Evidence level, applicability, indirectness, and limitations appear in separate columns and are visible beside each synthesized claim.
• The bibliography has passed a citation check.

If any item is missing, the folder is a working draft. Label it as such so no one mistakes it for a finished evidence product.