Muse for Researchers and Analysts: Source-Gathering, Synthesis Memos, and a Watch That Stays Current
A researcher and analyst working setup for Muse: gather sources with citations you can check, turn them into synthesis memos that separate verified facts from…
By Professor Muse8 min read
Research work repeats three jobs: gathering sources, turning them into something a reader can act on, and keeping the whole thing current after you publish or brief. Most of the time goes to the first job, and most of the risk sits in the second. A synthesis memo reads smoothly whether its claims are solid or not, which is exactly why the gathering has to be strict.
That is the work Muse is good at in a research seat. It reads widely, keeps the source next to the claim, and drafts the memo from the evidence file instead of from memory. It leaves the judgment calls to you: which sources count, which claims are proven, and which conclusions the evidence will not support yet.
The rule for this lesson: no claim without a source you can open, and no conclusion stronger than the evidence behind it. Every brief Muse writes separates verified facts from leads, and every lead says what would confirm it. That rule keeps a fast draft from becoming a confident error.
The short version
- Run three jobs on a rhythm: a source-gathering brief when a question opens, a synthesis memo when the evidence is in, and a weekly topic watch that reports only what changed.
- Every claim in a memo names its source, with a quote or a page reference where one exists. A claim without a source goes back as a lead, not forward as a finding.
- Keep verified facts and leads in separate sections. Mixing them is how an unconfirmed report becomes a fact in the next meeting.
- The failure mode to watch for is a polished synthesis built on a secondary summary. Trace the load-bearing claim to its primary source before you brief it.
The Setup: The Evidence Desk
Call the whole arrangement the Evidence Desk. I run it as a market analyst covering enterprise AI agents, with one standing question that never closes: what are the major vendors actually shipping, charging, and restricting, and what is only rumored? The desk has three jobs, and each one maps to a prompt below.
The source brief. When a question opens, Muse gathers the primary sources first: announcements, documentation, filings, and reputable reporting, each with a link and a short note on what it actually says. Secondary coverage gets labeled as secondary, so it never quietly stands in for the source.
The synthesis memo. Once the sources are in, Muse writes a memo with three sections: what the evidence supports, where sources disagree, and what is still unknown. Disagreements stay in the memo. A memo that hides a disagreement is a memo that will embarrass you in the question period.
The topic watch. Once a week, Muse rechecks the same sources and reports only what changed: new announcements, updated pages, and corrections. A watch that re-summarizes everything every week is a watch nobody reads. A watch that reports three changes is a watch that gets read.
| Job | Muse does | You do |
|---|---|---|
| Source brief | Gathers primary sources, logs each claim with its source, labels secondary coverage | Decides which sources count, drops weak ones, sets the question |
| Synthesis memo | Drafts findings, disagreements, and open questions from the evidence file | Checks the load-bearing claims, rewrites conclusions, signs the memo |
| Topic watch | Rechecks sources weekly, reports only what changed | Reads the changes, updates the memo, retires stale leads |
Before any of this runs, set the standing instructions from our memory guide: how you want sources cited, which outlets you treat as primary, and what counts as a lead rather than a finding. That instruction, stored once, is what keeps every brief in the same format. And the watch itself is a scheduled routine, the same shape as a morning briefing, with one extra rule: if nothing changed, the report says so in one line and stops.
The Evidence Desk builds on our guide to memory that compounds:
Prompt 1: The Source-Gathering Brief
Run this when a question opens, before anyone writes a conclusion about it.
Research [question, for example: what checkout and agent-access rules have major AI agent platforms announced this quarter?]. Gather primary sources first: official announcements, documentation, and filings. For each source, give me: the publisher, the date, the link, and one short note on what it actually says. Then list each claim you found with the source behind it. Label any claim that comes only from secondary coverage as a lead, and say what primary source would confirm it. Do not write conclusions yet. Brief only. I decide which sources count.
A brief that starts with conclusions is a brief that cherry-picks sources to fit them. A brief that starts with sources gives you a question you can actually answer.
Prompt 2: The Synthesis Memo
Run this once the evidence file exists, and only from that file.
Using only the sources in [evidence file or brief], write a synthesis memo on [question]. Structure it in three sections: findings the evidence supports, with the source behind each claim; disagreements between sources, with both sides stated fairly; and open questions, with what would answer each one. Quote the source where a claim turns on exact wording. If a claim in the brief has no source you can open, move it to the open questions section and say so. Do not send or publish the memo. Draft only. I check and sign it.
A memo with a disagreements section looks less tidy than one without it. It is also the memo that survives the first hard question, because the disagreement is already on the page instead of waiting in the audience.
Prompt 3: The Weekly Topic Watch
Run this on the same day each week, against the same source list.
Recheck the sources in [source list] for [topic]. Report only what changed since [last check date]: new announcements, updated pages, corrections, and retractions. For each change, give me the source, the date, and one sentence on what changed. If nothing changed, say so in one line and stop. Do not re-summarize unchanged sources. Do not add new sources without labeling them as new.
The watch works because it is boring. The week it reports three changes, you read all three. The week it re-summarizes forty sources, you read none of them.
What Can Go Wrong: The Polished Memo Built on a Summary
Here is the failure mode you will meet first. The synthesis memo comes back clean and well structured: findings, disagreements, open questions, all in your format. One finding carries the whole conclusion, and it cites a respected article. The article is real, the quote is accurate, and the conclusion follows from it. But the article is summarizing a vendor announcement, and the summary softened a condition the announcement stated plainly. The vendor’s rule applies only to one product tier, and the memo presents it as applying to all of them. The chain looks solid at every step except the one nobody checked: the step from the summary back to the source.
If that memo had gone to a client, the client would have planned around a rule that does not apply to their tier, and the correction would have arrived after the decision. The fix has two parts. First, Prompt 2 already requires a source you can open behind every claim, and you hold it to that: a claim whose only source is a summary gets treated as a lead until the primary source is in the file. Second, before you sign any memo, check the one load-bearing claim yourself: open the primary source and read the sentence the conclusion rests on. It takes a few minutes. That check is the research version of the rule in our failure modes lesson: a fluent draft is not evidence, and the verification step belongs to you, every time.
The same discipline applies to the brief itself. Our task brief guide covers the other half of it: a vague question produces a vague source list, so name the question, the time window, and the sources that count before Muse starts gathering.
Start With One Question
Do not build the whole Evidence Desk this week. Pick one open question, run Prompt 1, and read the brief the way a skeptical editor would: open three sources, check that each note matches what the source says, and strike anything you cannot open. If the brief survives that reading, run Prompt 2 on the surviving sources, then set the watch for the same source list next week.
Run it that way and Muse does what it is good at in research: the gathering, the first draft, the watching. You keep what was always yours: which sources count, which claims are proven, and what the evidence will not support yet. That gives you briefs you can defend, memos that survive questions, and a topic watch you actually read. The edge is a claim with a source behind it, every time.
