A final source check is the last pass where you confirm that every claim in your draft still points to a source you can open, read, and defend. It is not a reread of your prose and not a fresh literature review. It is a bounded audit: you verify that citations resolve, that each cited work actually supports the sentence it sits next to, that your search record is complete enough to describe, and that nothing in the reference list is a phantom.
The task usually arrives with a deadline attached. A thesis chapter is due to your committee, a course paper is closing, or a supervisor has returned a marked-up draft and you have a week to respond. The goal of this article is to give you a repeatable sequence for that week, with checkpoints you can actually finish, a worked example, and clear criteria for calling the check done.
What state are you actually starting from?
Before you open the draft, name your starting condition. The check looks different depending on which of these you recognize:
- A finished draft with a reference list you built by hand. Your main risk is citation drift: a source was swapped, a page number moved, or a claim was softened after the citation was added.
- A draft assembled across months, possibly with AI assistance. Your main risk is a reference that looks plausible but does not resolve to a real paper.
- A supervisor's feedback round. Your main risk is that new sentences were added in response to comments without new sources.
- A winter-break or January cleanup before a new project. Your main risk is an unorganized library, where you cannot tell which saved papers were actually cited.
Write one sentence describing your situation at the top of a scratch file. That sentence sets the scope of the check and stops it from expanding into a rewrite.
Step 1: Freeze the draft and build a claim-to-source map
Start by making the draft immutable for the duration of the check. Save a copy labeled with the date. Every edit from here goes into a working version, so you can always compare.
Next, build the artifact that drives everything else: a claim-to-source map. Go through the draft and list, in order, every sentence that carries a citation. For each one, record four things:
- The sentence, quoted exactly.
- The citation as written in the text.
- The full reference as it appears in your bibliography.
- A status column you will fill in later.
This map is the single most useful artifact in the whole process. It converts a vague feeling of "I should check my sources" into a finite list of rows. If your draft has forty citations, you have forty rows, and you can see your progress.
Checkpoint: You are ready to move on when every in-text citation has a matching row and every row has a matching bibliography entry. Mismatches found here are the cheapest ones to fix.
Step 2: Confirm that every reference resolves to a real paper
A reference that cannot be found is the most damaging kind of error, because it undermines trust in everything around it. Work through your bibliography and confirm that each entry corresponds to a paper that exists, with the authors, title, venue, and year you have written down.
For drafts where references were generated or reformatted in bulk, a BibTeX-level screen is the efficient move. TrueCite checks BibTeX against real academic databases to flag hallucinated references, which lets you catch phantom entries before you start reading them one by one. Run it on your .bib file, then reconcile the flagged entries against your claim-to-source map.
For entries that pass, still open the paper. A reference can resolve and still be the wrong paper for the sentence. Mark each row as resolved and relevant, resolved but mismatched, or unresolved.
Checkpoint: No row remains unresolved. Every resolved but mismatched row has either a corrected citation or a rewritten sentence.
Step 3: Test whether each source supports its sentence
This is the slowest step and the one most often skipped. For each row, open the source and find the passage that supports your sentence. Read it in context.
Ask three questions of every row:
- Does the source say this, or something narrower? If your sentence generalizes beyond what the source states, narrow the sentence.
- Is this the original report or a later summary? Where possible, cite the primary work.
- Does the source's method match your use of it? A finding from one setting may not transfer to the claim you are making.
This is where rigorous practice shows. NIH guidance on scientific rigor treats careful assessment of prior knowledge, strengths, weaknesses, and knowledge gaps as a core part of research practice, and that standard applies to how you represent each source in your own text. If a sentence overstates a source, the fix belongs in the sentence, not in a footnote.
When you need to re-read a source closely or ask a specific question of a paper you have already saved, Library QA lets you ask questions based on papers uploaded to My Library or added from search results, which is useful for confirming what a paper actually argues before you commit to a sentence.
Checkpoint: Every row is marked supported, needs narrowing, or needs replacement. No row is left blank.
Step 4: Reconstruct and document your search
If your paper describes how you found your sources — and any paper with a methods section should — the search record needs to match what you actually did. Cochrane guidance on searching for studies recommends planning searches from eligibility criteria, using 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.
You do not need to have followed a systematic-review protocol to benefit from that list. Use it as an audit of your own record:
- Eligibility criteria: Can you state, in one or two sentences, what made a source eligible for your paper?
- Sources: Which databases or search tools did you use? List them.
- Terms: Can you reproduce your key search strings?
- Deduplication: Did you remove duplicate records, including multiple reports of the same study?
- Screening: Can you describe how you moved from records to included sources?
- Rerun: Have you searched again recently enough that new work would have surfaced?
If your search record is thin, rebuild it now rather than describing a process you cannot support. Where you need to find or re-find papers, WisPaper search gives you a scholarly search interface to locate them, and you can save the results you keep into My Library so your included sources sit in one organized place.
If your paper is a systematic review or claims to follow reporting standards, PRISMA 2020 provides 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. Use the checklist as a structural audit of your methods section, not as a badge.
Checkpoint: You can write a methods paragraph that a reader could follow to reproduce your search, and every element in it is something you actually did.
Step 5: Reconcile the library against the reference list
By this point you have a claim-to-source map, a screened bibliography, and a search record. The last structural check is a three-way reconciliation between your saved papers, your cited references, and your draft.
Sort your saved papers into three groups:
- Cited and verified: appears in the reference list and passed Step 3.
- Saved but uncited: you read it but did not cite it. Decide whether it belongs in the paper or should be set aside.
- Cited but not saved: a reference in your list that you cannot locate in your library. These are the highest-risk entries and should be reopened first.
If you are working across a longer horizon — a thesis that continues past this submission, or a January project start — set up a Research Projects area so the sources and notes for this paper stay grouped and you can create a new project for the next one without mixing them.
Checkpoint: The cited but not saved group is empty, and every saved but uncited paper has a decision recorded.
A worked example: the January cleanup before a new project
Suppose you finished a course paper in December and your supervisor's feedback arrived in early January, alongside the start of a new project. You have two weeks before the next deadline and a reference list you assembled quickly.
Day 1 — Freeze and map. You save a dated copy of the submitted draft and build a claim-to-source map. It has thirty-one rows. Two in-text citations have no matching bibliography entry; you fix those immediately.
Day 2 — Resolve. You export your .bib file and run it through TrueCite to flag hallucinated references. Three entries come back flagged. You remove two and replace one with a paper you can open and read.
Days 3–5 — Support test. You work through the remaining rows in batches of ten. Four sentences overstate their sources and get narrowed. One cites a later summary where the original report is available, so you switch the citation.
Day 6 — Search record. You write the methods paragraph. Your original search was informal, so you rerun the key searches, deduplicate, and record the terms and sources you actually used. You save the included papers into My Library.
Day 7 — Reconcile. You sort saved papers into cited, uncited, and missing. Two cited references are not in your library; you reopen both and confirm they support their sentences.
Day 8 — Respond to feedback. You address the supervisor's comments, and every new sentence that makes a claim gets a row in the map and a verified source.
Day 9 — Final read. You read the draft against the map one last time and confirm the reference list matches the in-text citations exactly.
Day 10 — Handoff. You submit, and you open a new project for the January work so the next paper starts with an organized library rather than a scramble.
A reusable checklist you can copy
| # | Check | Artifact | Done when |
|---|---|---|---|
| 1 | Draft frozen | Dated copy | A working copy exists separate from the submitted version |
| 2 | Claim-to-source map built | Table with sentence, citation, reference, status | Every in-text citation has a row |
| 3 | References resolve | Screened bibliography | No row is unresolved |
| 4 | Sources support sentences | Status column filled | Every row is supported, narrowed, or replaced |
| 5 | Search documented | Methods paragraph | A reader could reproduce the search |
| 6 | Library reconciled | Three sorted groups | No cited but not saved entries remain |
| 7 | Feedback addressed | Revised draft | Every new claim has a verified source |
| 8 | Final read | Clean draft | Reference list matches in-text citations exactly |
How do you know the source check is finished?
A source check is complete when all of the following are true:
- Every in-text citation has a bibliography entry, and every bibliography entry is cited in the text.
- Every cited reference resolves to a real paper with the correct authors, title, venue, and year.
- Every cited sentence has been read against its source and marked as supported, narrowed, or replaced.
- Your methods paragraph describes a search you can reproduce.
- Your library contains every cited paper, and every uncited saved paper has a recorded decision.
- Any sentences added in response to feedback have their own verified sources.
If any of these is false, you are not finished. If all are true, further checking is unlikely to change the draft, and you should submit.
Where WisPaper fits in the sequence
WisPaper is an AI research assistant for finding, screening, organizing, and working with academic papers, and the steps above map onto it at specific points rather than throughout. You can start from the WisPaper homepage if you are new to the tool.
- Finding and re-finding sources: use WisPaper search when you need to locate a paper or rerun a search for your methods record.
- Screening references: use TrueCite on your BibTeX to flag hallucinated references before the support test.
- Organizing included sources: use My Library to save and upload the papers you keep.
- Checking what a paper argues: use Library QA to ask questions of papers already in your library.
- Keeping projects separate: use Research Projects when this paper's sources need to stay grouped apart from your next project.
For longer tasks that go beyond a single submission — a literature review that will feed a thesis chapter, for instance — Scholar Agent supports literature review, inspiration, and reproduction use cases. If your next step is generating research directions from your topics, background, references, and experimental constraints, Idea Discovery is built for that. And if you need to plan a reproduction from an uploaded PDF, PaperClaw handles research-paper reproduction planning.
Your next action
Open your draft now and create the claim-to-source map. Give it four columns — sentence, citation, reference, status — and fill in the first ten rows before you do anything else. That single artifact turns the final source check from an anxious reread into a finite task with a visible finish line.




