How to Recover the Logic of an Old Research Project

An old project rarely fails because the ideas were bad. It stalls because the reasoning that connected the question, the search, the screening decisions, and the draft has gone quiet. Recovering that logic means rebuilding an auditable trail from the artifacts you still have — search strings, saved records, screening notes, drafts, coauthor comments — and then re-deriving the decisions those artifacts imply. The goal is not a perfect reconstruction of what you once thought. It is a defensible account of why each step followed from the one before, so you can finish the manuscript, hand the project off, or restart the review without repeating work.

This guide gives you a sequence with checkpoints, a worked example, a reusable checklist, and clear rules for when to stop or escalate.

Before you rebuild anything, classify what you actually have. Most stalled projects fall into one of four states, and each needs a different first move.

• Draft exists, logic missing. You have a manuscript file but cannot explain how the included studies were chosen. Start with the methods section and work backward to the search.
• Search exists, decisions missing. You have saved searches or exported records but no record of why some were excluded. Start with the screening trail.
• Artifacts scattered. Search exports, notes, and drafts live in different places. Start by inventorying, not interpreting.
• Handoff. Someone else ran the original work and you inherited it. Start by reconstructing the eligibility criteria before touching the records.

Write one sentence naming the state and the single artifact you trust most. That sentence becomes your anchor: every recovered decision must be consistent with it, or you flag the conflict rather than smoothing it over.

Recovery is an evidence-gathering task first. Collect what exists and label each item by what it can prove.

Keep this inventory as a working document. It is the backbone of your method trail, and it tells you where the gaps are before you start filling them with inference.

Checkpoint 1: You can name every artifact you hold and state, for each, what it can and cannot prove. If an artifact type is entirely absent, note it as a gap rather than assuming its content.

Logic flows from the question. Before re-reading a single record, restate the research question and the eligibility criteria in plain language. Cochrane's guidance on searching for and selecting studies is useful here: searches should be planned from eligibility criteria, and the process should be documented. That ordering — criteria first, search second — is exactly the logic you are trying to recover.

Practical steps:

1. Write the question as a single sentence.
2. List the inclusion criteria as testable conditions (population, intervention or exposure, outcome, study design, date range, language).
3. List the exclusion criteria and, where possible, the reason each exists.
4. Compare this restatement against the draft's methods section. Where they diverge, mark the divergence rather than resolving it silently.

If the original criteria were never written down, derive a provisional set from the draft and the screening notes, and label it provisional in your working document. A provisional criterion is a hypothesis about the old logic, not a recovered fact.

Search strings are the most recoverable part of an old project because they leave traces. PubMed, for example, supports field tags, phrase searching, filters, Advanced Search, saved searches, and a Search Details view that reveals how terms were translated and combined. If the project used PubMed, that details view can show you how a query was actually interpreted, which is often more informative than the string you remember typing.

For each search you can locate, record:

• The database and interface.
• The exact string, including field tags and filters.
• The date run, if recorded.
• The number of records retrieved.
• Any deduplication step.

Cochrane's guidance supports combining free-text terms with subject headings, using multiple sources, documenting the process, deduplicating records, and rerunning searches before publication. Use those as checks against your reconstruction: if the recovered trail shows only free-text terms or a single source, you have found a gap to address before the manuscript is finished, not a reason to abandon the project.

If the original strings are lost, rebuild them from the eligibility criteria and label the rebuilt search as a new search with its own date. Do not present a rebuilt string as the original.

Checkpoint 2: Every search you rely on has a string, a source, a date, and a record count — or is explicitly marked as reconstructed.

Screening is where old projects lose the most logic, because exclusion reasons are often recorded only as a checkmark. Cochrane recommends screening in stages and linking multiple reports of one study. Apply that structure retrospectively:

1. Deduplicate the recovered record set and note how many duplicates were removed.
2. Title and abstract stage: for each excluded record, recover or assign a reason category.
3. Full-text stage: for each excluded full text, record the criterion it failed.
4. Link reports: group multiple reports of the same study so your counts reflect studies, not papers.

Where a reason was never recorded, you have two honest options: re-screen the record against your restated criteria and label the decision as a new one, or mark it as unrecoverable and exclude it from your reported counts. Choose one and apply it consistently.

This is also the stage where a literature search tool earns its place in the workflow. If you need to re-run or extend the original search to close gaps, WisPaper search provides a scholarly search interface for finding academic papers, and the public WisPaper Scholar Search page explains the entry point for finding papers. Use it to regenerate the record set you are missing, then log the new search alongside the old one so the trail stays continuous.

Now compare the manuscript to the trail you rebuilt. Read the methods section line by line and ask, for each claim, which artifact supports it.

• Does the stated number of records match the export?
• Do the stated inclusion criteria match the screening reasons?
• Does the stated search date match the saved search?
• Are all cited studies present in the included set?

Where the draft and the trail disagree, you have three resolutions: correct the draft to match the trail, correct the trail if you find a better artifact, or narrow the claim so it is true under both. Record which resolution you chose and why. That record is itself part of the method trail.

For citation-level checks, TrueCite screens BibTeX citations against real academic databases to flag hallucinated references — useful when an old bibliography contains entries nobody can now locate. Run it on the citation file before you finalize the reference list, and treat flagged entries as items to verify against the original source.

Checkpoint 3: Every quantitative claim in the methods section traces to a named artifact, or has been narrowed.

A two-author review on a clinical intervention stalled after the first draft. The search was run two years earlier; one coauthor has left; the screening spreadsheet has checkmarks but no reasons.

Inventory. The team holds a PubMed saved search, an exported record set, a screening spreadsheet, a draft manuscript, and email threads. No search date is recorded in the draft.

Question and criteria. The draft states the population and intervention but not the date range. The team restates the criteria and marks the date range as provisional, derived from the earliest included study.

Search trail. The saved PubMed search is recovered with its field tags and filters. The Search Details view shows how terms were combined, confirming the string. The run date is recovered from the export file's metadata rather than from memory, and logged as such.

Screening. The spreadsheet's checkmarks are converted into reason categories by re-screening a sample against the restated criteria. Records that cannot be re-screened are marked unrecoverable and excluded from reported counts. Multiple reports of the same trial are linked so the count reflects studies.

Draft reconciliation. The draft's record count does not match the export. The team corrects the draft to the export count and adds the deduplication step, which was never described.

Citation check. The bibliography is screened with TrueCite; two entries cannot be matched and are verified against the original sources before the reference list is finalized.

Result. The methods section now describes a search with a string, source, date, and count; a staged screening process with reasons; and a citation list that has been checked. The logic is recoverable because every step points to an artifact.

Recovery produces a pile of records, notes, and decisions. If they stay scattered, the logic will drift again. Consolidate the papers you are keeping into one organized place — My Library lets you save or upload papers and organize them in the WisPaper library. A single organized set makes the included studies auditable at a glance.

Once the set is assembled, you can interrogate it directly. Library QA lets you ask questions based on papers uploaded to My Library or added from search results. Use it to check whether the recovered included set actually supports the claims in the draft — for example, by asking which included papers address a specific outcome. Treat the answers as pointers back to the papers, and confirm each against the source before it enters the manuscript.

If the project is large enough to warrant a dedicated workspace, Research Projects is the WisPaper area for creating and managing a project. Keeping the recovered trail inside a project space makes the next handoff or revision far easier than reconstructing it a second time.

Checkpoint 4: The included set lives in one organized location, and every draft claim you checked has been confirmed against a source paper.

The reconstruction is ready when a reader who has never seen the project could follow the logic from question to conclusion without asking you a question. Test it against this checklist:

• The research question is stated in one sentence.
• Eligibility criteria are listed and testable.
• Every search has a source, string, date, and record count, or is marked reconstructed.
• Deduplication is described.
• Screening is staged, with reasons for exclusion.
• Multiple reports of one study are linked.
• Every quantitative claim in the draft traces to an artifact.
• Divergences between draft and trail are recorded with the resolution chosen.
• Citations have been checked against real sources.
• The included set is organized in one place.
• Gaps are labeled as gaps, not filled with inference.

If every box is checked, the logic is recovered well enough to finish the manuscript or hand off the project. If most are checked but a few gaps remain, narrow the affected claims and proceed. The trail does not need to be complete; it needs to be honest about what is missing.

Recovery can absorb unlimited time. Use these rules to decide when to stop or escalate.

Stop and finalize when the checklist is substantially complete and remaining gaps affect only claims you can narrow. Narrow the claim, record the gap, and move to writing.

Stop and re-scope when the recovered criteria no longer match the draft's question. This is not a recovery failure; it means the project's logic changed and the manuscript should follow the recovered criteria.

Escalate to a coauthor when a divergence between draft and trail cannot be resolved from artifacts alone — for example, when two screening decisions conflict and no record explains why. A short written query naming the artifact and the conflict is more useful than a general request to "check the methods."

Escalate to a new search when the recovered trail shows a structural gap: a single source where multiple were intended, free-text terms only where subject headings were needed, or a search that predates the project's own inclusion window. Run the new search, log it with its own date, and describe both searches in the methods.

Escalate to a fresh review only when the question, criteria, and included set cannot be reconciled at all. In that case the old project becomes background for a new one, and the recovered artifacts become its starting inventory.