A Client Briefing Contains Claims Nobody Can Trace: Running an Enterprise Provenance Audit

A briefing lands on the partner's desk the morning it is due. The argument is clean, the recommendations are confident, and three of the load-bearing claims have no visible source. One cites "industry data" with no year. Another paraphrases a study nobody on the team can locate. A third is a number that appears in two drafts but in no document. The memo is not wrong yet — it is unverifiable, which for a time-sensitive decision is the same practical problem.

This article walks through a provenance audit for that situation: rank the claims that actually carry the decision, recover the source records behind them, and reissue the briefing with explicit confidence labels. The example is a due-diligence memo prepared for a client considering an acquisition, but the same sequence applies to any evidence-backed report where a wrong claim would change a recommendation.

A claim without provenance cannot be checked, updated, or defended. When a client asks "where does this come from?" the team has no answer, and the recommendation loses authority even if it happens to be correct. The risk compounds in a time-sensitive memo because there is no slack to redo the analysis.

The repair is not to soften every sentence. It is to separate claims that drive the recommendation from claims that merely add color, then spend the remaining time on the ones that matter. That is the core discipline of a provenance audit: triage first, recover second, relabel last.

Picture the memo as it arrived. It recommends proceeding with the acquisition, and its argument rests on a handful of statements:

• Market claim: the target's segment is growing faster than the client's current segment.
• Technical claim: the target's process is more reproducible than the client's incumbent method.
• Risk claim: prior attempts at this integration failed for reasons the target has since resolved.
• Supporting claims: assorted context about competitors, timelines, and regulatory posture.

Only the first three change the recommendation. The supporting claims shape tone. A provenance audit that treats all of them equally will run out of time before it reaches the ones that matter.

Before searching for anything, classify each claim by what happens if it is wrong.

List every assertion that could be challenged, then tag each with two questions: does the recommendation depend on it, and would a contrary finding change the recommendation? Claims that answer yes to both are critical. Claims that inform but do not flip the decision are supporting. Claims that are decorative are contextual and can be cut or softened without loss.

Not all critical claims are equal. A claim that a single source can settle is cheaper to audit than one that requires synthesis across many studies. Rank critical claims by both decision weight and recoverability, and start with the ones that are both decisive and tractable. In the example, the technical claim is the best first target because reproducibility is a documented research property with a clear evidence trail.

Recovery means finding the document, study, or dataset that actually supports the sentence as written — not a document that supports a nearby idea.

The Cochrane Handbook's guidance on planning searches is useful here even outside clinical research: define what would count as eligible evidence before you search, rather than accepting whatever appears first. For the technical claim, that means deciding in advance that you need a study comparing reproducibility of the two methods, not a general article praising the target's approach.

Cochrane also recommends drawing on multiple sources and combining free-text terms with subject headings. In practice, that means searching the concept in more than one phrasing and more than one database before concluding a source does not exist. A single failed query is not evidence of absence.

This is where a scholarly search interface earns its place. Using WisPaper search to run the concept across phrasings lets the auditor see whether the claim has a real literature behind it or only a rhetorical one. For readers who want a public entry point before committing to a workflow, the WisPaper Scholar Search page describes the same finding task.

Cochrane's process guidance is explicit about deduplicating records, screening in stages, and linking multiple reports of one study. That last point matters in a provenance audit: a claim may appear to have three supporting sources when it actually has one study reported three times. Counting reports instead of studies inflates confidence.

Finding a source is not the same as validating a claim. The audit has to compare the sentence to the source.

NIH guidance on scientific rigor treats careful assessment of prior knowledge, strengths, weaknesses, and knowledge gaps as part of rigorous practice. Applied to a memo, that means asking whether the source addresses the same population, method, and time frame as the claim. A reproducibility finding from one laboratory setting may not transfer to the target's production process, and the memo should say so.

A study can be transparent about its methods and still have a narrow result. NIH guidance pairs transparent methods with validation and reproducibility as separate concerns. The auditor should record both: how well the source documents its approach, and how far its findings actually reach.

PRISMA 2020 is a reporting checklist for systematic reviews, not for client memos. But its structure — transparent reporting of search, selection, and synthesis methods — is a useful mirror. If the memo cannot describe how its evidence was found and selected, that gap belongs in the audit findings.

Confidence labels fail when they are adjectives. "High confidence" with no definition is decoration. Define each label by the evidence behind it.

• Verified: a specific source was located, and its scope matches the claim as written.
• Supported with limits: a source was located, but its scope, setting, or date narrows the claim; the memo states the limit.
• Unresolved: no source was recovered, or the recovered source does not support the sentence. The claim is removed, rewritten as a question, or flagged for the client.

A paragraph can mix a verified statement with an unresolved one. Labeling at the sentence level prevents a strong claim from lending borrowed credibility to a weak neighbor.

For each unresolved claim, note what was searched and what was found. This turns a dead end into a documented finding, which is itself useful to the client. It also prevents a later reviewer from repeating the same search.

A provenance audit produces a pile of papers, notes, and screening decisions. If that pile lives in scattered folders, the next reviewer cannot reconstruct the reasoning.

Saving the recovered papers into a single organized library makes the audit reproducible. My Library is the WisPaper area for saving or uploading papers and organizing them, which fits the step where the auditor consolidates what was recovered before relabeling claims.

Once the sources are collected, the auditor often needs to check a specific detail across them — for example, whether any recovered study reports a contrary result. Library QA supports asking questions based on papers uploaded to My Library or added from search results, which is the natural next move after the source set is assembled.

For a multi-week due-diligence effort, the audit itself becomes a workstream. The Research Projects area is where a team can create and manage that workstream rather than leaving the trail in email threads.

The corrected memo is not a weaker document. It is a document whose confidence is legible.

A claim labeled "supported with limits" should carry the limit in the sentence: the setting, the date, or the population where the evidence holds. A claim labeled "unresolved" should either be removed or converted into an open question with an owner and a deadline.

Attach a short provenance appendix: the claim register, the confidence label for each critical claim, and the source behind it. Clients rarely ask for this, but it changes the conversation from "trust us" to "check us."

Cochrane recommends rerunning searches before publication. The same discipline applies to a time-sensitive briefing: a final pass catches sources that appeared during drafting and confirms that nothing critical was missed.

In the example, the revised memo keeps its recommendation but changes its footing. The market claim is downgraded to "supported with limits" because the recovered source covers an adjacent segment. The technical claim is verified against a study whose scope matches the comparison, with the setting noted. The risk claim is marked unresolved and converted into a diligence question for the client's integration team.

The recommendation survives, but now it rests on claims the client can inspect. The memo also carries a short appendix listing what was searched and what was not found — which is often the most useful page in the document.

• Rank before you search. Decision weight, not prominence, determines where audit time goes.
• Treat a failed search as a finding. Document it; do not let it become a silent gap.
• Count studies, not reports. Multiple reports of one study are one source.
• Define confidence labels in advance. Adjectives without definitions do not survive review.
• Keep the trail in one place. An audit that cannot be reconstructed is not an audit.

For teams that run this cycle repeatedly, a longer research task can be handled through Scholar Agent, which fits literature review and related research work. And when a memo cites references in BibTeX form, TrueCite checks BibTeX against real academic databases to flag hallucinated references — a direct fit for the recovery step where a citation may not correspond to a real paper.