Shipping research outputs.

Three shapes, not a generic “research output” CORE

A huge amount of student research effort goes sideways because the student does not know what shape of document they are trying to produce. They end up with a hybrid — half memo, half brief, half outline — that does none of the three jobs well. Module 4 commits to three concrete shapes. Each has a purpose, a length, a structure, and a reader.

Shape 1: the scoping brief. One page. Five parts — question, reader-and-decision, answer-shape, out-of-scope, constraints (you wrote this in Lesson 4.2). Purpose: to define what the research task is before the research runs. Reader: often you, sometimes a teacher or parent you are scoping the task with. When to ship it: before retrieval starts. A scoping brief is the first document of almost every research project worth running. Even when you are the only reader, writing it is the step that prevents weeks of wasted work.

Shape 2: the fact-check memo. 400–800 words. Five sections — claim, verdict, evidence, confidence, what-would-change-your-mind (you wrote this in Lesson 4.3). Purpose: to answer one bounded question definitively. Reader: a decision-maker or a reader who needs a verdict rather than an overview. When to ship it: when a specific claim is blocking a decision or needs to be adjudicated. Fact-check memos are the shape most students have never written and will get the most professional mileage out of.

Shape 3: the synthesis brief. 1,000–2,000 words. Six-part plan, prose structured around the reader’s question, uncertainty held near the claim (you drafted this in Lesson 4.4). Purpose: to answer a larger question by synthesizing multiple claims and sources. Reader: someone who needs a complete, defendable answer to a bounded but non-trivial question. When to ship it: when a decision or a writing project needs more than a verdict — it needs the case.

These are not the only research-output shapes in the world. A literature review, a competitive teardown, an opinion column, an academic paper — all are legitimate shapes that fall outside these three. The course commits to three because three is what a student can ship well in Module 4, and because all three transfer outside the course. A student who can produce a clean scoping brief, a clean fact-check memo, and a clean synthesis brief has most of the research craft the working world asks of a serious junior.

Matching output shape to reader-and-decision CORE

Which shape fits which request? A short decision tree.

Is the purpose of the document to define the work rather than do it? → Scoping brief.

Is the purpose to answer one specific claim with a verdict? → Fact-check memo.

Is the purpose to answer a larger question with a reasoned, sourced case? → Synthesis brief.

Is the purpose something else — original reporting, a long literature review, an op-ed? → Outside this module’s commitments; pick the closest shape and extend, or use a different course.

A common student mistake is to reach for synthesis brief when a fact-check memo would do. If the reader is asking “is X true?” the right document is the one that says true / partially true / unclear / false — not a 1,500-word survey. Inverse mistake: reaching for a fact-check memo when the real question is broader than one claim. If the reader is asking “should we switch curricula?” a memo saying “true, curriculum A exists” does not serve them; they need the synthesis.

A useful habit: before you draft, write the reader’s actual question in their voice, in one sentence. If their question starts “Is it true that...” — fact-check memo. If their question starts “Which of these...” or “How should we...” — synthesis brief. If their question is “What exactly are we asking?” — scoping brief.

The final open-before-cite audit CORE

Before any brief ships, it gets one more pass. This pass is mechanical and non-negotiable and is short, bounded, per-brief work. It is the cost of the shipping rule.

Step 1 — build the citation list. Open your brief. List every source citation in the document. Number them. Do not trust memory — you are going line by line.

Step 2 — cross-check against source-list.md. For every citation in the brief, find the corresponding entry in your source list. Its status must be opened-and-confirmed. If any citation’s status is not-yet-opened or is missing from the list entirely, the citation is dropped from the brief before shipping. No exceptions. You either open the source now and add it to the list, or you remove the citation.

Step 3 — sample-read three citations. Pick three citations at random from the list. For each, open the source, find the specific sentence the brief uses it for, and confirm the source actually says what the brief claims. If any of the three fails, the entire brief goes back for another pass — not only the failing citation. A brief that failed one random sample is a brief whose other citations cannot be trusted without checking.

Step 4 — check thesis / body alignment. Read your thesis sentence. Read the first sentence of each body paragraph. Read the last paragraph. Is everything arguing toward the same place? If not, note the drift and revise.

Step 5 — run the synthesis-failure-mode pass. Read with the three failure modes (false voice, fact-free connector, single-source-as-consensus) in mind. Mark any sentence that fits a failure mode. Fix each one, even if it costs a little smoothness.

Only after all five passes is the brief ready. This is the research-agent equivalent of the "I have reviewed the agent's work" confirmation in Module 3 — the step that cannot be skipped no matter how pressed the deadline.

The shape of the capstone freeze CORE

Capstone freeze means: the brief is done, you have saved it in its final form, and you are not changing it for the rest of the course. This is your fourth frozen capstone artifact. It goes with:

• my-first-loop.md — from Module 1, the backbone of your agent-system design.
• The Module 2 workstation posture — your local/cloud choice and cost-ceiling documented.
• /capstone/directed-edit-log-v1.md — from Module 3, three real directed edits on a real codebase.
• /capstone/research-brief-log-v1.md — this module, three research outputs on a real topic.

The research-brief log has one file, structured as three sections. Each section is one complete output (scoping brief, fact-check memo, synthesis brief) plus its metadata: the prompt you wrote, the source list with confirmation marks, the agent session summary (or a link if the session is preserved), and a one-paragraph reflection on what you learned from that specific output.

A clean research-brief-log-v1.md is one a parent documenting credit can open, skim quickly, and see the shape of the student’s work. It is also one the student themselves can use as a reference in future research projects — “here is what a good fact-check memo looks like, by me.”

The Recipe Book and why your entries might end up in it CORE

Module 4 — like every module — depends on a set of Recipe Book entries that are dated and versioned and refreshed quarterly. The Module 4 entries today include running-a-research-agent-in-the-cowork-tab, running-a-research-agent-in-the-claude-code-cli (optional advanced), saving-sources-to-a-local-folder, and spotting-a-fabricated-citation. Each has a Last verified date, a Next review date, and a maintenance workflow.

The quarterly refresh cycle works like this. Every 90 days, someone — usually a course maintainer, sometimes a graduate student contributor — runs through each entry, re-tests it on the current tool versions, and either marks it fresh, patches it, or retires it. Entries that stop working (the tool changed in a breaking way, the tool no longer exists) are retired. Entries that changed cosmetically are patched. New tools get new entries.

You can contribute to this. When you finish your capstone and notice that a Recipe Book entry you used has a stale screenshot, a step that no longer matches the current UI, or a gap you worked around — flag it. The quarterly refresh relies on student feedback from people who ran the recipe last week. That is high-signal data. If you want to write an entry for a new research tool the Recipe Book does not yet cover, there is a contribution template at /recipe-book/_contribute.md.

This is not pretend participation. Recipe Book entries are the most obsolescence-prone layer of the course, and students currently running them are the best possible quality signal. The two-layer architecture of the course — durable Core Book, versioned Recipe Book — was built on exactly this assumption, and it works only if the Recipe Book gets refreshed.