Research Project Closeout: Preserve Sources, Decisions, and Search History

A research project is not finished when the last analysis runs. It is finished when someone else—a coauthor, a reviewer, a successor in the lab, or your future self—can reconstruct what you searched, what you kept, why you kept it, and how the final claims were built. Closeout is the discipline of turning a working project into an auditable record.

This guide treats closeout as a set of decisions and artifacts rather than a single checklist. It is organized so you can enter at whatever stage your project is actually in: still searching, mid-screening, drafting, or already responding to reviewers. The examples draw on systematic-review practice, where the trail is most explicit, but the same logic applies to any manuscript that depends on a body of literature.

Closeout is the point where your working materials become a record. The record needs to answer four questions: which sources you considered, how you found them, which ones you excluded and why, and which decisions shaped the final manuscript.

Systematic-review methodology makes this concrete. The Cochrane Handbook recommends planning searches from eligibility criteria, drawing on multiple sources, combining free-text terms with subject headings, documenting the process, deduplicating records, screening in stages, linking multiple reports of one study, and rerunning searches before publication. PRISMA 2020 supplies a reporting checklist, expanded explanations, an abstract checklist, and flow-diagram templates for original and updated reviews, including transparent reporting of search, selection, and synthesis methods.

You do not need to run a systematic review to borrow the structure. You need the same three layers: a source layer (what you found), a decision layer (what you kept and why), and a search layer (how you found it and when). Closeout is complete when all three layers are consistent with each other and with the manuscript.

Most closeout failures come from starting in the wrong place. Before you build anything, locate your project on this diagnostic.

Ask three questions to place yourself. First, can you name the criteria that determined inclusion? Second, can you reproduce your last search and get a comparable set? Third, can you explain any single exclusion without reopening the paper? If any answer is no, that layer is your starting point.

The source layer is the set of records you actually considered. Its quality depends on how you searched and how you handled duplicates.

Cochrane's guidance is to combine free-text terms with subject headings and to search multiple sources, then deduplicate records before screening. PubMed supports field tags, phrase searching, filters, Advanced Search, saved searches, Similar Articles, citation lookup, and links to full text when available; its search details reveal how terms were translated and combined. That translation view matters at closeout because it lets you document not just what you typed but how the database interpreted it.

Google Scholar documents searching by author and title, date sorting and date limits, Cited by and Related articles, citation export, library access links, email alerts, and ways to locate alternative versions of a work. Those features are useful for building and extending a source set, but the export and alert records are what make the set auditable later.

A practical source-layer artifact contains, for each database or source: the exact query, the date run, any limits applied, the number of records returned, and the number removed as duplicates. If you are using WisPaper to find papers, the scholarly search interface is where that discovery step happens; save the queries you run so the log reflects real searches rather than a reconstruction.

The search layer is the most commonly lost artifact, because searches feel disposable while you are running them. They are not. A search you cannot rerun is a claim you cannot defend.

Record, at minimum:

• The source searched and the interface used.
• The full query string, including field tags and any subject headings.
• Filters and date limits.
• The date the search was run.
• The number of results and how they were exported.

Cochrane recommends rerunning searches before publication. That recommendation has a closeout consequence: your search log should have a final entry dated close to submission, not only entries from the start of the project. If the rerun changes the record set, the change belongs in the log and, where relevant, in the manuscript's methods.

For projects where the search itself is the deliverable, the Scholar Agent area supports longer research tasks, including literature review work. Use it as part of the search workflow, then keep the resulting queries and record sets in your own log so the trail survives independently of any single session.

The decision layer is where most projects are thinnest. Screening usually happens quickly and informally, and the reasons for exclusion live in someone's memory.

Cochrane recommends screening in stages and linking multiple reports of one study. Both practices generate decisions worth recording. Staged screening means a record can be excluded at title level, abstract level, or full-text level, and the stage matters for reporting. Linking reports means several records may map to one study, so your unit of decision is the study, not the record.

A workable decision log has one row per record or study with: identifier, screening stage reached, decision, reason category, and the person who decided. Reason categories should be short and stable—eligibility criteria, wrong population, wrong design, duplicate report—so they can be counted later.

PRISMA 2020's flow-diagram templates are built around exactly this kind of counting: records identified, records screened, records excluded, and studies included. Even if you are not writing a systematic review, sketching that flow for your own project exposes where the decision trail has gaps.

The most damaging closeout error is drift: the manuscript cites a source set that no longer matches the record set, or the methods describe a search that was never actually run in that form.

Reconcile in three passes. First, check that every cited work appears in your source layer; a citation with no corresponding record is either a late addition that needs logging or an error. Second, check that the methods section describes the searches in your log, including the final rerun. Third, check that any exclusion you describe in the text matches a reason category in the decision log.

Citation integrity deserves its own pass. BibTeX entries accumulate errors and, in generated bibliographies, can include references that do not correspond to real publications. TrueCite checks BibTeX against real academic databases to flag hallucinated references, which makes it a useful step when your bibliography was assembled or edited under time pressure. Run it before submission, not after a reviewer questions a reference.

Consider a coauthored review that has come back from peer review with a request to justify the search strategy and explain why several apparently relevant papers were excluded.

The closeout sequence runs as follows. Pull the search log and confirm the final rerun date; if no rerun exists, run one now and record it. Rebuild the deduplicated record set from the exports. Reconstruct the decision log for the papers the reviewer named, using the original eligibility criteria rather than post hoc reasoning. Reconcile the manuscript's methods and flow description against those artifacts. Then verify the bibliography.

For the verification step, TrueCite's BibTeX citation screening fits naturally after the decision log is settled, because you want the reference list stable before you check it. For the reconstruction step, if your team stored PDFs in a shared library, Library QA lets you ask questions based on papers uploaded to My Library or added from search results—useful when you need to re-check what a specific excluded paper actually reported without rereading it end to end.

The handoff package is the reader-facing layer. It should be short enough to read in one sitting and complete enough to answer follow-up questions without you.

Include:

• A one-page project summary with the research question and eligibility criteria.
• The search log, with the final rerun clearly marked.
• The deduplicated record set and the decision log.
• The manuscript version and the bibliography as submitted.
• A short open-questions note: what you would search next, which exclusions were borderline, and which decisions were judgment calls.

That last item is often omitted and is often the most valuable. A successor needs to know where the trail is firm and where it required judgment.

If the project will continue under a new phase, the Research Projects area is where a new project can be created so the next phase starts with its own workspace rather than inheriting an unlabeled folder. For teams that keep PDFs centrally, My Library is where papers can be saved or uploaded and organized, which makes the source layer easier to hand over as a coherent set.

Closeout is ready when an informed outsider can pass this test without asking you questions.

1. Reproduction. They can rerun your final search from the log and obtain a comparable set.
2. Reconciliation. Every cited work maps to a record, and every described exclusion maps to a reason category.
3. Currency. The search log contains a rerun dated close to submission.
4. Attribution. Each decision in the log has a named decider and a date.
5. Readability. The handoff summary explains the project without requiring the full record set.

If any item fails, fix that layer before archiving. A partial closeout that is honest about its gaps is more useful than a complete-looking package with a broken search log.

If you are still searching. Write eligibility criteria first, then start the search log on day one. Every query gets an entry. This is the cheapest possible time to build the trail.

If you are mid-screening. Stop and reconstruct the decision log for records already screened, using stable reason categories. Then continue screening with the log open.

If you are drafting. Reconcile the methods section against the search log before you polish prose. Drift caught at drafting is cheap; drift caught at review is expensive.

If you are responding to reviewers. Treat each reviewer comment as a query against your artifacts. If an artifact cannot answer it, that gap is the finding, and the fix belongs in the record.

If you are handing off. Build the handoff summary last, from the other artifacts, and include the open-questions note.

For readers who want a general entry point before choosing a workflow, the WisPaper homepage introduces the assistant across finding, screening, organizing, and working with academic papers. Readers working in Chinese can start from the Chinese homepage.