Most researchers do not lose a literature search because the database stopped working. They lose the why. Six months after a screening pass, the folder is still there, but the connection between the eligibility criteria, the query, the excluded records, and the final citation list has quietly disappeared. When a reviewer asks for the search strategy, or a coauthor asks why a known paper was excluded, the audit trail is gone.
This article is for academic researchers and faculty who are revising a manuscript, closing a project, protecting an evidence trail, or starting a new review. It treats archiving as a deliberate handoff from active work to a stable record, not as deleting files. You will learn how to diagnose what you actually have, decide what must be preserved, build an artifact that survives a future you, and know when the archive is finished.
A reliable archive serves three jobs. First, it lets anyone reconstruct which searches produced which records. Second, it lets you explain screening decisions without reopening the raw data. Third, it lets you update a review later without re-deriving the entire method. Cochrane's guidance captures the core requirement: searches should be planned from eligibility criteria, documented, deduplicated, screened in stages, and updated before publication. An archive is what makes that documentation durable.
What kind of archive am I really creating?
Before organizing anything, identify your starting state. Most researchers fall into one of four situations, and each needs a different archive.
Manuscript submission or revision. The search supported a paper that is now moving toward publication. The archive needs to prove that the method was transparent and reproducible, and that no eligible record was silently dropped.
Project closure. The grant ended, the lab handoff is happening, or the student is graduating. The archive needs to let another person take over without a live conversation.
Evidence preservation. The search informed a policy brief, a clinical decision, or a registered protocol. The archive needs to freeze the evidence trail at a specific date.
Starting a new review. You are about to begin a systematic search and want a clean structure from the start. The archive you build now is the one you will hand off later.
Your starting state determines which artifacts matter most. A manuscript revision needs a query log and a screening audit. A lab handoff needs the deduplication record and the inclusion notes. A new review needs the eligibility criteria documented before the first query runs.
Which searches are worth archiving?
Not every exploratory lookup deserves a formal archive. Apply a simple filter before you do any work.
Archive the search when at least one of these is true:
- It produced records that appear in a manuscript, grant, or report.
- It is part of a registered protocol or systematic review.
- It will need to be updated, rerun, or audited later.
- Someone other than you may need to explain or reproduce it.
- It cost real time to design, deduplicate, or screen.
Skip formal archiving for quick definitional lookups or one-off inspiration searches. The goal is not to document every query ever typed. It is to protect the links between method, records, and decisions that would be expensive to reconstruct.
One useful starting practice is to keep the archive tied to the scholarly record itself. When you save papers that came from a decision-relevant search, organize them deliberately in My Library under a named collection that corresponds to the manuscript or review, not under a generic "misc" folder. The collection name becomes part of the audit trail.
What does a complete search archive contain?
A defensible archive contains five layers, each capturing a different part of the method.
1. The method plan. This is the eligibility criteria and the decision of which sources to search. Cochrane's guidance emphasizes planning from eligibility criteria and using multiple sources. Write down the population, intervention or exposure, comparator, and outcome before the query exists.
2. The queries. Each database query, the date it ran, the platform or database name, and any filters applied. A saved screenshot of the query interface can help, but the query text itself is the core record.
3. The deduplication record. The count of records before and after removing duplicates, and how you identified duplicates. Cochrane explicitly calls this out: deduplicate records and link reports of the same study.
4. The screening record. The staged decisions: titles and abstracts screened, full texts screened, included, excluded, and the reason for each exclusion. Linking multiple reports to one study matters here, because two records can describe the same underlying work.
5. The update plan. When the search was last run and when it should run again. Cochrane recommends updating searches before publication, so the archive needs a dated last-run marker.
A useful mental model for the final record is the persistent identifier used in scholarly infrastructure. The DOI Foundation describes a DOI as a persistent identifier that resolves through the DOI system and can remain stable when the location or descriptive metadata of an object changes. Applied to search archives, the insight is practical: the archive should have a stable identifier and a stable enough location that someone can still cite it when the folder moves. This is why naming conventions and a dated manifest matter more than drag-and-drop tidiness.
How do I freeze the search record without losing the live workspace?
Archiving does not mean breaking your active workflow. The trick is to create a stable snapshot at logical checkpoints while the live workspace keeps moving.
A practical checkpoint structure looks like this:
Checkpoint 1: after eligibility criteria are set. Save a short method note with the criteria and intended sources. This locks the plan before any query runs.
Checkpoint 2: after each database search. Export the query string, hit count, and date into a run log. Do not wait until the end of screening.
Checkpoint 3: after deduplication. Record the pre- and post-dedup counts and the tool or rules used. This number will appear in your PRISMA-style flow and must be reproducible.
Checkpoint 4: after title and abstract screening. Freeze the list of records moving to full text, with reviewer assignments if there are multiple screeners.
Checkpoint 5: after full-text screening. Freeze included and excluded records, and attach an exclusion reason to every excluded full text. This is the layer most often missing in manuscript revisions.
Checkpoint 6: before submission or handoff. Bundle everything into a dated archive folder, write the manifest, and mark the last search date.
Between checkpoints, keep working normally. The archive is a series of dated snapshots, not a daily journal.
Which folder structure survives a coauthor or a year?
A flat pile of exports fails the moment someone asks "which version did we use?" A survivable structure separates raw outputs from decisions and never edits the raw outputs in place.
Use this structure inside a single project archive folder:
review-archive/
01-protocol/
02-searches/
03-deduplication/
04-screening-ti-ab/
05-screening-full-text/
06-final-included/
07-manifest.md
Keep the numbered folders read-only after the project closes. Any correction goes into a new dated copy, never over the original export. The manifest is the only file that changes as the archive matures.
In the manifest, answer five questions in plain text:
- What research question does this archive serve?
- When was the last search run?
- Which databases or platforms produced the records?
- Where are the final included records listed?
- Who is responsible for the archive after handoff?
For manuscript revisions where citation integrity matters, consider an additional check on the BibTeX or reference file itself. Once you have a reference list that feeds the manuscript, run it through TrueCite, which checks BibTeX against real academic databases to flag hallucinated references. The archiving moment is exactly when a phantom citation, introduced during a rushed revision, becomes costly to repair.
What should a worked example look like?
Imagine a faculty member revising a scoping review after reviewer comments. The reviewers asked for the full search strategy and questioned why a well-known paper was absent.
The researcher's eligibility criteria were: adults with the target condition, any intervention that included the reported component, any comparator, and outcomes measured at least six weeks after baseline. The search ran in three databases in January and was updated in August.
The archive process took one afternoon.
First, they wrote a protocol note in 01-protocol/ restating the eligibility criteria and listing the three databases. This answered the first reviewer question immediately.
Second, they reconstructed the run log in 02-searches/ by exporting the three query strings and recording the hit counts and dates. The August update was dated, which satisfied the Cochrane requirement to update searches before publication.
Third, they recorded the deduplication step. Crossref exposes registered scholarly metadata through its REST API, including DOI, title, author, publication, funder, license, and relation fields when deposited, so they used a DOI-based deduplication rule to merge records with the same registered DOI. The dedup record showed that one study had two reports, which they linked in the screening log.
Fourth, they rebuilt the exclusion log. The absent paper appeared in the full-text screening level with the exclusion reason "wrong outcome timing." That single line resolved the reviewer's concern without reopening the entire dataset.
Finally, they placed the included papers in My Library under a collection named for the review, ran the BibTeX through TrueCite to catch reference errors, and wrote the manifest. The revision response included the manifest path, the query strings, and the exclusion log reference.
The result is an archive that answers reviewer questions directly instead of generating new work under deadline.
How do I hand off an archive to someone else?
A handoff succeeds when the next person can answer method questions without contacting you. Test this by asking someone unfamiliar with the project to open the archive and report back the number of included studies and where the search strategy lives.
For lab handoffs, the most important file is the manifest. A new graduate student should be able to open 07-manifest.md and learn the five manifest answers in under two minutes. The raw folders remain sealed; the manifest is the navigation layer.
For coauthor handoffs, share the archive folder plus a one-paragraph handoff note that points to the three things most likely to be needed in revision: the query strings, the dedup counts, and the full-text exclusion reasons.
When the project will produce future work based on the same evidence base, link the archive to a living project space. In Research Projects, a researcher can create a new project that holds the protocol, the derived questions, and the next steps connected to the archived search record. The archive stays frozen; the project becomes the active workbench.
How do I know the archive is ready?
Use explicit completion criteria. A ready archive passes all five checks below.
| Check | Pass condition |
|---|---|
| Protocol present | Eligibility criteria and source plan are written in a versioned file |
| Queries dated | Every query has a date, a database name, and the exact query text |
| Dedup reproducible | Pre-dedup count, post-dedup count, and dedup rule or tool are recorded |
| Screening auditable | Every full-text exclusion has a reason; reports of the same study are linked |
| Update documented | The last search date is recorded, and the next update window is stated if relevant |
If any check fails, the archive is not ready for submission or handoff. Fix the gap before moving on.
For a stronger readiness test, ask one question of the archive as a whole: can a colleague reconstruct the included record set from the method files alone? If the answer requires a conversation with you, the archive is still a personal memory, not a durable record.
What stays in the archive versus what stays readable?
There is a useful distinction between two storage goals within an archive. The first is frozen evidence, which must remain untouched: raw exports, dated snapshots, and the original query strings. The second is working memory, which benefits from staying queryable: the included papers, the screening notes, and the questions that connect them.
Freeze the evidence layer. Keep the working memory layer connected to a tool that can answer questions about the saved papers. When a coauthor asks "what did the included studies actually show about follow-up length?", the fastest answer comes from a queryable set of the included papers rather than from reopening frozen exports. In Library QA, a researcher can ask questions based on papers uploaded to My Library or added from search results, which turns the archived included set into a reasoning resource without disturbing the frozen record.
The discipline is simple: never delete or edit the frozen layer, and freely query the working layer. The archive remains intact while the work stays accessible.
Reusable archive checklist
Copy this checklist into the manifest at the start of the project and tick items as they complete.
Before searching
- Eligibility criteria written and versioned
- Source list finalized
- Archive folder structure created
- Record-linking rule chosen for multiple reports of one study
During searching and screening
- Each query exported with date, database, and count
- Dedup counts and rule recorded
- Title/abstract screening decisions exported
- Full-text screening decisions exported with exclusion reasons
- Multiple reports of the same study linked
At closing
- Last search date recorded
- Final included list frozen
- BibTeX or reference file checked
- Manifest written with the five manifest answers
- Handoff owner named
Readiness test
- A colleague can identify the included record set from the archive alone
- Five-point readiness table passes completely




