A 90-Minute Evidence-Gap Rescue Plan Before Submission

You have a deadline, a rubric, and a draft with soft spots. Some claims rest on a single citation, some have none, and you are not sure every reference is real. Ninety minutes is enough to triage those gaps, retrieve targeted evidence, verify what you cite, revise the weakest passages, and make a defensible submission decision — provided you work in a fixed sequence and honor stop rules.

The plan below divides the time into five blocks: claim triage (15 minutes), targeted retrieval (25 minutes), source verification (20 minutes), revision (20 minutes), and final checks plus the submission decision (10 minutes). Each block ends with a checkpoint. If a checkpoint fails, you escalate or stop rather than drift into the next block.

NIH guidance on scientific rigor treats careful assessment of prior knowledge, strengths, weaknesses, and knowledge gaps as a core part of rigorous research practice. That ordering matters under deadline pressure: you cannot search efficiently until you know which claims are actually weak. Cochrane's search guidance likewise begins with eligibility criteria — you define what evidence would count before you go looking. A rescue plan that starts with searching produces a pile of papers and no decision about which claims they are meant to support.

So the first block is diagnostic, not productive in the visible sense. You are building a short, ranked list of claims that need evidence, and deciding in advance what would satisfy each one.

Open your draft and mark every sentence that asserts a fact, a comparison, a mechanism, or a trend. Then sort each marked claim into one of four buckets:

• Supported and verified — has a citation you have already checked.
• Supported but unchecked — has a citation you have not confirmed.
• Unsupported — needs a source.
• Unsupportable in the time available — you cannot find or verify evidence within this session.

Rank the unsupported and unchecked claims by how much the rubric depends on them. A claim carrying a results paragraph or a core argument outranks a background sentence. Write, next to each of the top five claims, the kind of evidence that would settle it: a primary study, a review, a methods reference, or a definition.

Checkpoint 1: You should leave this block with no more than five claims in play and a one-line evidence target for each. If you have more than five, cut to the five the rubric weights most. If you cannot name what evidence would satisfy a claim, move it to the unsupportable bucket now rather than later.

Retrieve only for the claims on your shortlist. Cochrane recommends combining free-text terms with subject headings and drawing on multiple sources, so build each query from the claim's key concepts rather than typing the claim as a sentence. Keep a running log: claim, query used, source consulted, date.

For scholarly search, use WisPaper search to find academic papers for each shortlisted claim, and save the ones you intend to cite into My Library so retrieval and organization stay in one place. If a claim needs a broader literature pass rather than a single paper, the Scholar Agent area supports longer research tasks including literature review.

As records accumulate, deduplicate and screen in stages, as Cochrane describes: title and abstract first, then full text for the survivors. Link multiple reports of the same study so you do not count one finding twice.

Checkpoint 2: For each shortlisted claim, you should have either a candidate source or a clear "not found in this window" note. If three or more claims still have no candidate, stop retrieving and go to Block 5's escalation rule — more searching will not fix a shortlist that was wrong.

A candidate source is not evidence until you have confirmed it exists, says what you think it says, and is the version you intend to cite. Work through your candidates in rubric order.

For each source: confirm the reference details, read the passage you plan to rely on, and check that the claim you make is narrower than or equal to what the passage supports. If your draft cites BibTeX entries, run them through TrueCite, which checks BibTeX against real academic databases to flag hallucinated references. If you need to interrogate papers you have already saved, Library QA lets you ask questions based on papers uploaded to My Library or added from search results.

Cochrane also recommends rerunning searches before publication. In a 90-minute window, treat that as a final spot-check in Block 5 rather than a full rerun.

Checkpoint 3: Every claim you plan to keep must have a verified source, or be rewritten to match what the source actually supports. Any reference that fails verification is removed, not left in place. If more than two references fail, go directly to Block 4 and cut the claims they supported.

Revise in this order: delete, narrow, then cite.

1. Delete claims in the unsupportable bucket. Removing a sentence is faster and safer than defending it.
2. Narrow surviving claims so their wording matches the verified evidence — swap "proves" for "is consistent with," drop universal quantifiers, and remove causal language the source does not support.
3. Cite at the point of the claim, with the verified reference.

If your paper reports a search or selection process, PRISMA 2020 supplies a reporting checklist and flow-diagram templates covering search, selection, and synthesis methods. Use it as a completeness check on what you have already done, not as a new task to start now.

Checkpoint 4: Read each revised paragraph aloud. If a sentence still outruns its source, cut it. If the revision has removed so much that a core section no longer stands, that is an escalation signal, not a reason to reinstate weak claims.

A student has a course paper due at 17:00 and a rubric that weights argument (40%), use of evidence (30%), and citation accuracy (20%), with 10% for presentation. The draft has eleven uncited claims. Triage keeps five: two in the argument section, two in the evidence section, one in the introduction. The introduction claim is dropped immediately because the rubric barely rewards it.

Retrieval finds candidate sources for three of the four remaining claims. Verification confirms two and fails one reference that turns out not to match any real record. Revision deletes the failed claim, narrows one surviving claim from a causal statement to an associational one, and cites the two verified sources. The fourth claim has no source at the checkpoint.

At the final decision, the student has a paper whose argument section is intact, whose evidence section is thinner but accurate, and whose citation list contains only verified entries. The submission decision is to submit with a short note to the instructor identifying the one claim that was removed for lack of verifiable evidence. That is a defensible submission: the rubric's citation-accuracy weight is protected, and the gap is disclosed rather than hidden.

Run four checks in order:

1. Every in-text citation appears in the reference list, and every listed reference is cited.
2. No reference remains that failed verification.
3. Spot-check one search per major claim, following Cochrane's rerun-before-publication practice at the smallest useful scale.
4. Confirm the paper still meets the rubric's structural requirements.

Then make one of three decisions:

• Submit. All kept claims are verified, the reference list is clean, and remaining gaps are disclosed.
• Submit with a disclosure note. One or two claims were removed or narrowed for lack of verifiable evidence; say so briefly to the instructor.
• Escalate. A core section cannot stand without the removed claims. Contact the instructor or supervisor before the deadline with a specific description of the gap and what you can deliver.

• Stop retrieving when your shortlist is covered or when three claims remain unfound. Additional searching past that point consumes revision time without improving the paper.
• Stop verifying a source after two failed attempts to confirm it; remove the reference and the claim it supported.
• Escalate if revision would remove the evidence base of a section the rubric weights heavily, or if a supervisor's feedback requires evidence you cannot obtain in the window.
• Never reinstate an unverified reference to fill a gap. An empty citation slot is a disclosed gap; a fabricated one is a different category of problem.