Full-text screening fails quietly. A reviewer reads a PDF, decides it does not fit, and moves on—but the reason lives only in their memory, the exclusion is never logged against the right criterion, and six weeks later a coauthor asks why a plausible study was dropped. The decision existed; the evidence of it did not. The fix is not more diligence in the moment. It is a screening method that produces a decision record for every full-text you touch, so that the trail survives the reviewer, the revision, and the handoff.
This guide treats full-text screening as a small production system: inputs, decision rules, artifacts, checkpoints, and completion criteria. It is written for researchers revising a manuscript, closing a project, preserving an evidence trail, or starting a new review. The method draws on established review practice—Cochrane's guidance on planning searches from eligibility criteria, screening in stages, and documenting the process; PRISMA 2020's reporting checklist and flow-diagram templates; and NIH's emphasis on transparent methods and reproducibility—and shows where WisPaper's tools fit at specific steps.
Why do full-text decisions disappear?
Most lost decisions come from four ordinary gaps, not carelessness.
- The criterion was never operational. "Relevant to the topic" cannot be applied consistently across reviewers or across weeks. Cochrane's guidance starts from eligibility criteria and builds the search from them; if the criteria are vague, every downstream decision inherits the vagueness.
- The decision was made but not recorded. Stage-by-stage screening only works if each stage leaves a trace. A verdict with no reason, date, or reviewer attached cannot be audited or reversed.
- Records were never deduplicated or linked. Cochrane notes that the same study can appear in multiple reports. If you screen reports rather than studies, you can exclude one report of an included study and never notice.
- The search moved after screening. Cochrane recommends rerunning searches before publication. A rerun that is not reconciled against earlier decisions creates two incompatible screening states.
Each gap has a matching artifact. The rest of this article builds those artifacts in order.
What should exist before the first full-text is opened?
Full-text screening is the second stage, and it depends on a clean first stage. Before you open a single PDF, confirm four things exist.
- A written eligibility protocol. State inclusion and exclusion criteria in operational terms—population, design, date range, language, outcome—so that two reviewers reading the same paper would reach the same verdict. Cochrane's recommendation to plan searches from eligibility criteria is the anchor here: the criteria drive the search, and the search defines what reaches full text.
- A documented, deduplicated record set. Cochrane recommends multiple sources, free-text terms combined with subject headings, and deduplication of records. Your full-text queue should be a deduplicated list, not a raw export.
- A title/abstract screening outcome. Cochrane describes screening in stages. Full-text screening should begin only for records that passed the earlier stage, and that pass should itself be recorded.
- A decision log schema. Decide now what every full-text decision will capture: record ID, study ID, reviewer, date, verdict, exclusion reason tied to a numbered criterion, and a short note. This schema is the artifact that prevents loss.
If you are assembling the candidate pool, WisPaper search provides a scholarly search interface for finding academic papers, and the WisPaper Scholar Search page explains the entry point for finding papers. Neither replaces your protocol; they feed the queue your protocol defines.
How do you set up a decision log that survives revision and handoff?
A decision log is a table, not a narrative. Build it before screening and keep one row per full-text record. A reusable structure:
| Field | What it captures | Why it matters |
|---|---|---|
| Record ID | Stable identifier from the deduplicated set | Lets you trace back to the source record |
| Study ID | Groups multiple reports of one study | Supports linking reports per Cochrane guidance |
| Reviewer | Who made the call | Enables a second reviewer to check or arbitrate |
| Date | When the decision was made | Orders decisions across search reruns |
| Stage | Title/abstract or full text | Keeps staged screening auditable |
| Verdict | Include, exclude, or pending | Distinguishes open questions from closed ones |
| Criterion | Numbered inclusion/exclusion rule applied | Ties the verdict to the protocol |
| Note | One or two sentences of reasoning | Explains edge cases to coauthors and reviewers |
| Linked reports | Other records for the same study | Prevents double-counting or accidental exclusion |
Two conventions make the log durable. First, never leave a verdict blank—use "pending" with a note about what is missing. Second, tie every exclusion to a numbered criterion from the protocol, so the reason is checkable against a document that does not change mid-screening.
PRISMA 2020's flow-diagram templates give you the reporting shape this log should eventually feed: counts at identification, screening, and inclusion. If your log is structured to produce those counts, the diagram becomes a summary rather than a reconstruction.
What does a full-text decision actually require?
A defensible full-text decision has three parts: the criterion applied, the evidence from the paper that satisfies or fails the criterion, and the verdict. Write all three into the log row.
Work criterion by criterion, not impression by impression. For each full-text record:
- Confirm identity. Check whether this record is a new study or another report of one already in the log. Cochrane's guidance on linking multiple reports of one study applies directly: link, do not duplicate.
- Apply criteria in a fixed order. Put the most decisive, easiest-to-check criterion first (for example, design or date range) so obvious exclusions close quickly and consistently.
- Record the first failing criterion. For exclusions, log the criterion that failed, not every criterion. This keeps reasons comparable across reviewers.
- Write the note for the next reader. The note should let a coauthor understand the call without reopening the PDF.
- Mark pending explicitly. If a full text is unavailable or a detail is ambiguous, log "pending" with the specific question to resolve.
This is where a library helps. Saving the full-text set in My Library keeps the papers you are screening organized in one place, so the record IDs in your log correspond to papers you can reopen. When you need to interrogate the saved set—for example, to check how several included papers describe a method—Library QA lets you ask questions based on papers uploaded to My Library or added from search results.
How should two reviewers reconcile disagreements?
Disagreement is data, not friction. The goal is a recorded resolution, not a silent override.
- Screen independently first. Each reviewer fills their own rows against the same numbered criteria.
- Compare at the criterion level. A disagreement about the verdict usually traces to a disagreement about which criterion applies or how it is operationalized.
- Resolve into one row. Record the final verdict, the resolving reviewer, and a note explaining the resolution. Keep the original rows or a note of the disagreement so the reconciliation is visible.
- Fix the protocol, not just the case. If a disagreement reveals an ambiguous criterion, amend the protocol with a dated note and re-check earlier decisions against the clarified rule.
This is the checkpoint where most decision loss happens in practice: reviewers agree verbally, one updates the log, and the reasoning vanishes. Make the resolution a written field.
How do you keep the search and the screening in sync?
Cochrane recommends rerunning searches before publication. A rerun is only safe if you can tell new records from old ones and reconcile new decisions against the existing log.
- Timestamp the original search. Record the date, sources, and terms used.
- Deduplicate the rerun against the existing set before screening anything new.
- Screen new records through the same staged process—title/abstract first, then full text—using the same numbered criteria.
- Reconcile. If a new record is another report of an already-included study, link it. If it changes a prior verdict, update the row with a dated note rather than overwriting it.
- Update the counts. Your PRISMA-style flow counts should reflect the rerun.
For longer reconciliation work—such as rebuilding the search narrative for a revision—Scholar Agent supports longer research tasks in WisPaper's literature review, inspiration, and reproduction use cases. Use it to organize the work around your log, not to replace the log.
What does a complete screening trail look like? A worked example
Consider a team revising a review manuscript before resubmission. Their protocol defines five numbered criteria. Their full-text queue has 40 deduplicated records.
- Setup. They create the decision log with the fields above and save the 40 full texts in My Library so each record ID maps to an openable paper.
- First pass. Two reviewers screen independently. Reviewer A excludes record 17 under criterion 3 (design). Reviewer B includes it. At reconciliation, they discover criterion 3 did not specify how to treat mixed-methods designs. They resolve record 17 as excluded, log the resolution, and amend criterion 3 with a dated clarification.
- Linking. Record 29 turns out to be a second report of the study in record 12. They link the reports under one study ID rather than counting two studies.
- Rerun. Before resubmission, they rerun the search, deduplicate against the original 40, and find six new records. Two are additional reports of included studies and are linked; four go through staged screening.
- Reporting. Because every row carries a verdict, criterion, and date, the team can populate a PRISMA 2020 flow diagram from the log and describe selection transparently in the methods, consistent with PRISMA's reporting checklist.
- Citation check. Before submission, they verify the reference list. TrueCite checks BibTeX against real academic databases to flag hallucinated references—a final integrity pass on the citations that survived screening.
The example is unremarkable on purpose. A trail that works looks boring: every decision has a row, every exclusion has a criterion, every rerun has a date.
What checkpoints tell you screening is on track?
Set checkpoints at fixed points rather than at the end.
- Before full text: protocol written, criteria numbered, queue deduplicated, title/abstract outcomes recorded.
- After the first ten full texts: sample the log. Can a coauthor read any row and understand the decision without the PDF? If not, tighten the note standard now.
- At each disagreement cluster: confirm resolutions are written, not verbal.
- Before any rerun: confirm the original search is timestamped and the log is current.
- Before reporting: confirm counts reconcile across stages and linked reports are grouped under study IDs.
These checkpoints map to the rigor NIH describes: careful assessment of prior knowledge, transparent methods, and reproducibility as core parts of research practice. A screening log is one of the most concrete reproducibility artifacts a review produces.
When is full-text screening finished?
Screening is complete when all of the following are true:
- Every full-text record has a verdict—include, exclude, or a documented pending with a resolution plan.
- Every exclusion cites a numbered criterion from the current protocol.
- Every decision has a reviewer and a date.
- Multiple reports of the same study are linked under one study ID.
- The final search rerun is reconciled against the log, with new records screened and prior verdicts updated by dated note where needed.
- Stage counts reconcile and can populate a PRISMA 2020 flow diagram without reconstruction.
- A coauthor can audit any row without reopening the source PDF.
If any item fails, the gap is specific and fixable—which is the point of building the log first.
How do you hand off or archive the trail?
A screening trail outlives the screening. Two handoff scenarios matter most.
Coauthor or successor takes over. Hand over the protocol (with dated amendments), the decision log, the deduplicated record set, and the search timestamps. The successor should be able to resume at the pending rows without re-reading included papers. Keeping the project organized in Research Projects gives the team a shared place to create and manage the project around these artifacts.
Project close or manuscript revision. Archive the log alongside the manuscript materials so that a reviewer's question about an exclusion can be answered from the record. If the revision requires new literature work, Scholar Agent can support the longer literature review tasks, and if the project turns toward reproducing a method from an included paper, PaperClaw supports research-paper reproduction planning from an uploaded PDF and reproduction instructions.




