Anatomy of a Research Task
What this is. A classification exercise. Six short information requests. For each: (1) what kind of task is it, (2) how would you scope it if a research agent is the right tool, and (3) which tool would you reach for first.
Why it matters. Not every information request is a research task. Some are summaries, some are primary-document reads, some should not be handed to an agent at all. This worksheet builds that instinct before Lesson 4.2.
The four categories
| Code | Category | When to use it |
|---|---|---|
| R | Research agent | The question depends on sources outside the model, you want a range, and you want to defend the answer. Scope → retrieve → triangulate → synthesize. |
| S | Summary | The model's training is plenty. A summary answer is fine; retrieval adds nothing. |
| P | Primary doc | There is one document, and the right move is to read it — possibly with an agent helping you parse. |
| H | High-stakes | Cost of a confident wrong answer is high and verification is hard. The agent finds the right specialist; it does not answer alone. |
The six tasks
For each, circle or write the category, then fill in the scoping sketch (only if R), first-choice tool, and a one-sentence reason.
“What are the three main arguments for a year-round school calendar, and the three main counter-arguments, using sources published since 2023?”
Category: R / S / P / H
Scoping sketch (only if R) — question, answer shape, source typesFirst-choice tool: Cowork Claude Code Either N/A
One-sentence reason“Define mitochondrial DNA.”
Category: R / S / P / H
Scoping sketch (only if R)First-choice tool: Cowork Claude Code Either N/A
One-sentence reason“My mom is picking between two laptop models for our family homeschool — a ThinkPad X1 and a MacBook Air M3. Which is the better fit for our setup, and why?”
Category: R / S / P / H
Scoping sketch (only if R) — question, answer shape, source typesFirst-choice tool: Cowork Claude Code Either N/A
One-sentence reason“Summarize the plot of Great Expectations.”
Category: R / S / P / H
Scoping sketch (only if R)First-choice tool: Cowork Claude Code Either N/A
One-sentence reason“What are the medication interactions of drug A and drug B, and should I take them at the same time?”
Category: R / S / P / H
Scoping sketch (only if R)First-choice tool: Cowork Claude Code Either N/A
One-sentence reason“This letter is from my grandfather, written in 1974. What is he arguing in it?”
Category: R / S / P / H
Scoping sketch (only if R)First-choice tool: Cowork Claude Code Either N/A
One-sentence reasonDeliverable
A filled-in copy of this worksheet, saved as anatomy-of-a-research-task-<your-name>.md in your Module 4 topic folder. Keep it — the scoping sketches for Tasks 1 and 3 are the closest pre-written models you will have for your own scoping brief in Lesson 4.2.
Answer key — resist until you have filled in the worksheet
Task 1 — R. A comparison of contested positions with a date constraint is classic research-agent work. Cowork if standalone; Claude Code if the output feeds a larger paper.
Task 2 — S. A definition request. A summary from any competent model is plenty. Using a research agent here confuses “research” with “work that looks impressive.”
Task 3 — R. The strong zone: specific, bounded, depends on current pricing and feature pages, and a real decision is downstream. Cowork is the friendlier first choice.
Task 4 — S. A plot summary of a canonical novel is solidly in the model's training. Agent-level retrieval adds nothing.
Task 5 — H. High-stakes medical. The agent's role is to summarize what is known and tell you to talk to a pharmacist or physician. Do not act on an agent-written brief here.
Task 6 — P. A personal letter you possess is a primary document short enough to read yourself. An agent can help parse a phrase; it cannot substitute for reading.
Safety norm — carries forward
You do not cite what you have not opened. Even in Lesson 4.1, before any brief is drafted, the habit begins: the student — not the agent — is the one who attests the work.
Print this page. Use it.