November 9, 2026

Your Coauthor Added Citations Without Saving the Sources: A Repair Workflow for Traceable Reports

You open the deck the night before the client review. The decision memo reads cleanly, the recommendation is sharp, and the footnotes look authoritative. Then you notice the problem: a coauthor added a dozen citations du

Written byWisPaper TeamAI Research Workflow Team
Editorial illustration for Your Coauthor Added Citations Without Saving the Sources: A Repair Workflow for Traceable Reports

You open the deck the night before the client review. The decision memo reads cleanly, the recommendation is sharp, and the footnotes look authoritative. Then you notice the problem: a coauthor added a dozen citations during the final pass, but no PDFs, no DOI list, and no notes came with them. The references exist as strings in a document, not as sources you can open, verify, or hand off.

This is a provenance failure, not a formatting problem. The citations may be perfectly real, or some may be misattributed, or a few may point to works that do not exist in the form described. You cannot tell from the document alone. What you can do is run a disciplined repair: diagnose what you actually have, decide which claims need which level of evidence, rebuild the source record, and mark confidence explicitly so the deliverable is honest about what it rests on.

The workflow below is written for enterprise research, analysis, and knowledge teams producing client decks, decision memos, team handoffs, and project closeouts. It treats citations as artifacts with provenance, not decoration.

What exactly broke when the sources went missing?

A citation without a saved source breaks three things at once.

First, it breaks identity. A reference string like an author, year, and title is a claim about a specific work. That work has a registered identity — typically a DOI — and bibliographic fields such as title, authors, venue, and date. Crossref exposes publisher-deposited scholarly metadata through its REST API, including titles, authors, venues, dates, identifiers, licenses, funding, ORCID and ROR identifiers, abstracts when deposited, and post-publication updates from trusted sources. That metadata is how you confirm the work is what the citation says it is.

Second, it breaks locatability. A DOI can be resolved through doi.org to the current destination registered for that identifier. Resolution helps you locate a work, but the bibliographic fields and the source text still need to be checked for identity and claim support. In other words, a working link is necessary but not sufficient.

Third, it breaks claim support. Even a correctly identified paper may not actually say what the memo claims. The citation string tells you nothing about whether the sentence above it is supported by the source text.

So the failure has layers. Your repair has to address all three, in order, or you will produce a document that looks sourced but still cannot be defended.

Diagnose the starting state before you touch the document

Before editing a single footnote, classify what you have. Most missing-source situations fall into a few recognizable states, and each needs a different response.

State 1: Citation strings only

The document contains author-year or numbered references with no accompanying files, links, or notes. This is the most common state after a fast coauthoring pass. You have claims and labels but no source record.

State 2: Partial identifiers

Some citations include a DOI or a URL; others do not. Partial identifiers are useful but uneven. A DOI gives you a resolution target and a metadata lookup path. A bare URL may point to a landing page, a preprint, or something that has moved.

State 3: Sources exist but are not attached to claims

A shared drive has PDFs, but nobody recorded which PDF supports which sentence. This is often worse than State 1 because it creates false confidence: the team believes the sources are "saved" when they are only stored.

State 4: Sources attached but unverified

Files and links exist and are mapped to claims, but nobody checked identity or claim support. This is the state where a quick verification pass pays off most.

Write the state down. It determines how much work the repair takes and what you tell stakeholders in the meantime.

Decide what level of evidence each claim actually needs

Not every sentence in a client deck deserves the same evidentiary weight. Trying to fully source every line wastes effort; sourcing none of them creates the problem you are fixing. Make an explicit triage decision.

A practical split:

  • Load-bearing claims — the numbers, market assertions, and causal statements that drive the recommendation. These need identified sources with checked claim support.
  • Contextual claims — background that shapes the reader's understanding but does not carry the decision. These need at least a locatable source.
  • Team assertions — judgments, estimates, and internal observations. These need to be labeled as such, not dressed in citations.

Record the triage in a simple table: claim, tier, current citation, source status, confidence. That table becomes the artifact you work from and, later, the artifact you hand off.

Rebuild the source record: identity first, then claim support

With triage done, rebuild the source record in two passes. Do not combine them; conflating identity and claim support is how teams reintroduce the original error.

Pass one: confirm each work exists and is what the citation says

For every load-bearing citation, establish the work's identity. Use the DOI where one exists and resolve it to confirm the current destination. Pull the deposited metadata to check title, authors, venue, and date against the citation string. Where a DOI is missing, search by author and title. 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 — useful when you need to find the correct record or an accessible version.

If you are working inside WisPaper, the scholarly search interface is where you locate papers and confirm you have the right record before saving anything. For a public explanation of the search entry point, the Scholar Search page describes finding academic papers.

Pass two: confirm the source supports the sentence

Open the source text and read the passage the claim depends on. Confirm the claim is stated, not inferred from the title or abstract alone. Where the claim is a paraphrase, check that the paraphrase does not strengthen the original. Where the claim is a number, confirm the number and its conditions.

This pass is where most silent errors surface. A citation can be perfectly real and still be attached to the wrong sentence.

Save what you verified so the next handoff does not repeat the failure

Verification that lives only in your browser history is not a source record. Save the works you confirmed and organize them so the next person can find them.

In WisPaper, My Library is where you save or upload papers and organize them. Use it as the durable store for the sources behind the deliverable, not as a scratch space. Pair it with the claim table so each saved paper maps to specific claims.

For team deliverables that span multiple workstreams, Research Projects gives you a project area to hold the work. Create the project at the start of the repair so the source record and the deliverable stay together.

A useful convention: name saved items so they match the claim table rows, and record the DOI and the verification date alongside each one. That way a reader six months later can see what was checked and when.

Screen the reference list for citations that do not resolve

When a coauthor adds citations quickly, some may be fabricated, garbled, or assembled from fragments of real works. Before you spend time chasing every reference, screen the list.

TrueCite checks BibTeX against real academic databases to flag hallucinated references. If your reference list can be exported as BibTeX, run it through TrueCite to separate references that need investigation from those that resolve cleanly. Treat the output as a triage signal: flagged entries go to the front of the verification queue, and clean entries still get the claim-support check in pass two.

This is the highest-leverage step when the citation count is large and the deadline is close. It converts an open-ended worry into a prioritized list.

Ask questions against the saved sources instead of re-reading everything

Once the verified papers are in your library, you can interrogate them directly rather than re-reading each one from the top.

Library QA lets you ask questions based on papers uploaded to My Library or added from search results. Use it to check whether a specific claim appears in the saved sources, to compare how two sources describe the same phenomenon, or to locate the passage behind a paraphrase. The Library QA area is the place to do that.

Keep the discipline: a QA answer points you to a passage, and you still confirm the passage supports the sentence in the deliverable. The question is a locator, not a verdict.

Assign explicit confidence levels to every load-bearing claim

A repaired deliverable should tell the reader how much weight each claim can bear. Confidence levels make that visible without burying it in footnotes.

A workable three-level scheme:

  • Verified — the work's identity is confirmed and the source text supports the claim as written.
  • Located, support unconfirmed — the work is identified and locatable, but the claim-support check is incomplete or the source text was inaccessible.
  • Team assertion — the claim reflects internal judgment or estimate and is labeled as such, with no citation attached.

Put the level in the claim table and, where the audience needs it, in the deliverable itself. A decision memo that marks two claims as team assertions is more credible than one that dresses them in citations nobody can open.

Rebuild the deliverable and the handoff artifact together

The corrected deliverable is not just the deck with fixed footnotes. It is the deck plus the source record plus the claim table. Build them as one package.

For a client deck, that means a reference appendix with resolved identifiers and a note on verification status. For a decision memo, it means a short provenance section stating which claims are verified and which are team assertions. For a team handoff or project closeout, it means the claim table, the saved library, and a short note on what was screened and what remains open.

If the repair is part of a larger research effort, Scholar Agent supports longer research tasks in the areas of literature review, inspiration, and reproduction. Use it when the repair reveals a gap that needs new literature work, not as a substitute for verifying what is already cited.

Know when the repair is actually finished

The repair is ready when a reader who was not in the room can trace every load-bearing claim to a saved, identified source and see its confidence level. Test it with a simple walkthrough:

  1. Pick three load-bearing claims at random.
  2. Follow each citation to the saved source.
  3. Confirm the identity fields match the citation.
  4. Read the passage and confirm it supports the sentence.
  5. Check that the confidence level is stated.

If all three pass, the package is in good shape. If any fail, you have found the next item for the queue. This walkthrough is also the fastest way to onboard a reviewer who needs to sign off.

Transferable lessons for the next coauthoring cycle

Missing sources are usually a process gap, not a character flaw. The coauthor added citations because the document needed them; nobody had agreed on where sources would live. Close that gap before the next cycle.

  • Agree on a source store before drafting. Decide where verified papers live and who maintains the mapping to claims.
  • Cite only from the store. A citation should be added by pointing at a saved, identified source, not by typing a reference string.
  • Triage claims early. Decide which claims are load-bearing so verification effort goes where it matters.
  • Screen reference lists before deep verification. Flagging unresolvable entries first saves hours.
  • Carry confidence levels into every handoff. The next person should inherit the claim table, not just the document.

These lessons apply whether the deliverable is a client deck, a decision memo, a team handoff, or a project closeout. The artifact changes; the provenance discipline does not.

FAQs

Start with identity. If the citation includes a DOI, resolve it through doi.org and check the deposited metadata against the citation string. If there is no DOI, search by author and title. For a whole reference list, export it as BibTeX and screen it with TrueCite, which checks BibTeX against real academic databases to flag hallucinated references. A resolved identifier confirms the work exists; it does not confirm the claim is supported, so follow with a claim-support check.