How to Build a Source-Checked Research Memo for a Client

A source-checked research memo is a short client-facing document in which every substantive claim can be traced to a specific source, and every source carries an explicit confidence level. The work is less about writing and more about building a defensible chain: eligibility criteria, a documented search, a screened evidence set, a claim-to-source map, and a confidence label the client can act on.

This guide walks through that chain as a sequence with checkpoints. It uses a worked example — a decision memo supporting a client deck — and ends with a reusable checklist and clear stop rules. Where a WisPaper capability fits a step, it is linked at the point you would actually use it.

A memo is source-checked when three conditions hold:

1. Traceability. Each claim in the memo points to a named source, and the reader can follow the pointer without asking you.
2. Documented selection. You can show how sources were found and why some were excluded, not just which ones survived.
3. Calibrated confidence. Each claim is labeled with how strongly the evidence supports it, so the client knows where to push back.

Search methodology guidance supports this structure. The Cochrane Handbook 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. Those practices translate directly into memo artifacts: a criteria sheet, a search log, a deduplicated source list, and a final re-check before the memo ships.

Most weak memos fail before drafting. Diagnose where you are:

• No criteria yet. You have a topic but no definition of what counts as in-scope evidence. Fix this first; everything downstream depends on it.
• A pile of PDFs, no log. You have sources but cannot reconstruct how you found them. You will need to rebuild a search record or narrow the memo's claims to what you can document.
• Claims without pointers. The draft reads well but individual sentences have no source. Expect a claim-by-claim audit pass.
• Sources without confidence labels. You have citations but no sense of which are strong. Add a labeling pass before the client sees anything.

Write one sentence describing your current state and the gap. That sentence determines which checkpoint below you start from.

Before searching, write two short artifacts:

• The client question, in one sentence, phrased so it could be answered yes, no, or with a bounded recommendation.
• Eligibility criteria: what kinds of sources count (study types, date ranges, populations, geographies), and what disqualifies a source.

Cochrane's guidance to plan searches from eligibility criteria is the anchor here: the criteria drive the search terms, not the reverse. If you search first and define scope later, you will spend the audit pass arguing with yourself about which sources belong.

Checkpoint: A colleague who has never seen the project can read your criteria and predict roughly which sources you will include. If not, tighten the criteria.

Use more than one source, and record what you did. Cochrane recommends multiple sources and combining free-text terms with subject headings; database documentation shows how to execute that in practice. PubMed supports field tags, phrase searching, filters, Advanced Search, saved searches, Similar Articles, citation lookup, and links to full text when available, and its search details reveal how terms were translated and combined. 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.

For the memo, the search log should capture: the databases or tools used, the exact query strings, date limits, filters applied, and the date the search ran. That log becomes an appendix the client can inspect.

If you are working inside WisPaper, the scholarly search interface is where you run paper searches, and the Scholar Search entry page explains the finding-papers capability for readers new to the tool. Save the query strings you use there into your log as you go — reconstructing them later is the most common source of audit pain.

Checkpoint: Someone else could rerun your search from the log alone and land on a substantially similar source set.

Cochrane recommends deduplicating records and screening in stages. In memo terms:

1. Deduplicate the raw results so the same work appearing in multiple databases is counted once.
2. Screen titles and abstracts against your eligibility criteria.
3. Screen full texts for the survivors.
4. Link multiple reports of the same study so the memo cites the study, not each report as independent evidence.

Record counts at each stage. A simple funnel — records found, duplicates removed, screened out, full-text excluded with reasons, included — is enough, and it signals rigor to the client without requiring a methods essay.

As papers survive screening, park them somewhere stable. In WisPaper, My Library is where you save or upload papers and organize them, which keeps the included set separate from the raw search results.

Checkpoint: Every excluded source has a one-line reason tied to a criterion. "Didn't seem relevant" is not a reason.

This is the core artifact. Create a table with one row per substantive claim in the memo:

Rules that keep the map honest:

• One claim per row. Compound sentences hide unsupported halves.
• Quote or paraphrase the supporting passage in the notes column so the audit pass does not require reopening every PDF.
• Label confidence explicitly. Use a small fixed scale — for example High / Moderate / Low — and define what each means for this memo before you start labeling.
• Flag single-source claims. A claim resting on one study is a different risk category from one resting on a body of work, and the client should see which is which.

If you need to interrogate the included papers directly — for instance, to confirm what a study actually reports before assigning confidence — Library QA supports asking questions based on papers uploaded to My Library or added from search results. Use it to check your reading of a source, then record the confirmed passage in the map.

Checkpoint: Read the memo's claims column alone. Each row should be a sentence you would defend in a client meeting.

Citation errors are a distinct failure mode from evidence errors. A reference can point to a real paper but contain a wrong year, a mangled author list, or a DOI that resolves elsewhere. Before the memo ships, verify each reference against a database record.

If your references are managed as BibTeX, TrueCite checks BibTeX against real academic databases to flag hallucinated references — a useful screen for catching entries that do not correspond to an actual paper. Pair it with manual spot-checks of author, year, and venue against the publisher record.

Checkpoint: Every reference resolves to a real, correctly attributed work. Zero unresolved entries.

With the claim map built, drafting becomes assembly. Structure the memo so the client can act:

• Bottom line up front. The recommendation or answer, with its overall confidence.
• Key findings, each carrying its confidence label and a pointer to the claim map.
• What would change our view. The evidence that would move a Low to a High, or reverse the recommendation.
• Appendix: search log, screening funnel, full claim-to-source map, reference list.

Confidence labels should reflect the evidence you actually assembled: number of independent sources, directness to the client's population or context, consistency across sources, and study type. A claim supported by one indirect study is Low regardless of how confidently it is written.

Checkpoint: No sentence in the body asserts something absent from the claim map. If drafting produced a new claim, it goes into the map and gets a source before it stays.

Situation. A client is deciding whether to fund a program, and your team must deliver a decision memo that will be condensed into three slides for a board deck.

Criteria. The memo will consider peer-reviewed studies and systematic reviews from the last decade, in the client's sector, in comparable economies. Opinion pieces and vendor white papers are excluded.

Search. You run queries in two databases plus a scholarly search tool, logging exact strings, filters, and run dates. You save included papers to a library so the set is stable.

Screening. You deduplicate, screen titles and abstracts, then full texts, recording counts and exclusion reasons. Two reports of the same study are linked and cited once.

Claim map. The memo makes six substantive claims. Four rest on multiple consistent sources (Moderate or High). One rests on a single study in a neighboring population (Low). One — the cost estimate — has no eligible source at all, so it is cut from the memo and flagged to the client as an open question rather than asserted.

Verification. References are checked against database records; BibTeX entries are screened for hallucinated references.

Delivery. The memo leads with the recommendation at Moderate confidence, names the single-study claim as the weakest link, and states what evidence would raise confidence. The three deck slides carry the same labels, so nothing strengthens in translation from memo to slide.

Handoff. The claim map, search log, and screening funnel ship with the memo so a colleague can update it next quarter without redoing the search from scratch.

Use this before any client-facing delivery:

• Client question written in one sentence
• Eligibility criteria documented and applied
• Search log complete: tools, query strings, filters, run dates
• Multiple sources used; free-text and subject headings combined where the database supports it
• Records deduplicated; screening done in stages with counts
• Multiple reports of one study linked and cited once
• Claim-to-source map covers every substantive claim
• Confidence scale defined and applied to every claim
• Single-source claims flagged
• Every reference verified against a database record
• Search rerun before publication to catch new evidence
• Memo, map, log, and funnel packaged together for handoff

Cochrane's recommendation to rerun searches before publication belongs on this list: a memo built on a search from months ago may miss evidence that changes a confidence label.

Know when to pause rather than push through:

• Stop if a load-bearing claim has no eligible source. Cut the claim or reframe the memo around what the evidence supports. Do not soften the language and keep the claim.
• Escalate if the recommendation flips under a single source. That fragility is itself a finding; tell the client the decision hinges on one study rather than presenting false stability.
• Stop if screening counts cannot be reconstructed. Rebuild the log before delivery, or narrow the memo's scope to the sources you can document.
• Escalate if references fail verification. An unresolvable reference is a blocker, not a formatting issue.
• Stop if the client's question has drifted. If the memo now answers a different question than the one in Step 1, renegotiate scope before continuing.

Some client memos grow: the question needs a literature review, or the client asks for a reproduction plan for a key study. At that point, move the work into a project structure rather than stretching a single memo. WisPaper's Research Projects area is where you create and manage a project, which keeps the search log, source set, and claim map together for the next deliverable. For longer literature-review-style tasks, the Scholar Agent area covers literature review, inspiration, and reproduction use cases.