literature-scan

name	literature-scan
description	Use when starting research on a new topic - produces a one-page map of the canonical 5-10 sources and the major positions
kind	workflow
audience	anyone
ai-tools	any
complexity	guided
time	60-120 min
version	1.0.0
source	bundled

Literature Scan

What this does

Maps an unfamiliar topic by finding the canonical 5-10 sources, identifying the major positions/camps/findings, and producing a one-page brief that you can navigate from. Avoids the "drowning in tabs" problem.

When to use

• Starting research on a topic you don't know well
• Preparing to write about, present on, or interview someone about a topic
• Choosing what to read deeply (after this scan)
• Building a literature review (this is the first stage)

What you bring (Inputs)

• The topic, narrowed to a single sentence
• The audience (yourself / a memo reader / a paper reviewer) — drives depth
• Time budget (1 hour for a quick scan, half-day for a thorough one)

What you get (Output)

A one-page brief with: scope, 5-10 canonical sources (with one-line summaries), the 2-4 major positions or camps, the consensus and the disagreements, and the gaps you noticed.

How it works (Steps)

This is a workflow.

Stages

Stage 1: Define scope precisely

Write the topic in one sentence. If you can't, the scope is too broad. Examples:

• Bad: "Climate policy"
• Good: "Carbon border adjustment mechanisms and their impact on developing-country exporters, 2020-2025"

Stage 2: Find the canonical sources

For 60-90 minutes, hunt for the 5-10 sources everyone in this area cites. Signals:

• Cited by multiple recent reviews
• Authored by people who appear repeatedly
• The seminal paper that established the field/term
• The most recent comprehensive review

Capture: title, author, year, one-line summary, why it matters.

Stage 3: Identify the camps

Read the abstracts/introductions of all sources. Group them. Are there 2-4 distinct positions, methodologies, or findings?

Examples: "Pro vs anti," "Empirical vs theoretical," "Three different mechanisms proposed," "Two regional perspectives."

Stage 4: Note consensus and disagreement

What does almost everyone agree on? What is actively contested? Where does new evidence point?

Stage 5: Note your gaps

What questions remain unanswered in what you read? What surprised you? What didn't you understand?

These become your reading list for deep-dive.

Stage 6: Write the brief

One page (or slightly more). Sections: scope, sources, camps, consensus, disagreement, gaps.

Checkpoints

• After Stage 1: scope fits in one sentence
• After Stage 2: 5-10 sources, no fewer (gap in canon) and no more (procrastination)
• After Stage 3: at least 2 distinct camps named
• After Stage 6: brief fits on one page; you could hand it to someone else and they'd know the lay of the land

Loop-back conditions

• If Stage 2 turns up 0-2 sources: scope is too narrow or you're using the wrong terminology — go back and re-search with synonyms
• If Stage 2 turns up 50+ sources: scope is too broad — narrow it
• If Stage 3 finds no camps: either everyone agrees (rare) or you didn't read enough — re-do

Quality bar

• Each source has a one-line summary written in your own words
• "Camps" are described, not just listed
• Disagreement is named specifically (not "they disagree on details")
• Brief is one page — if longer, it's a literature review, not a scan

Variations

• Pre-interview scan: scope around what the interviewee has said publicly
• Pre-decision scan: scope around the decision options
• Hot-topic scan: prioritize 2024-2026 sources

Example

Input: "What are the main approaches to evaluating large language model alignment in 2025?"

Output (one-page brief):

• Scope: ...
• 8 canonical sources: 3 RLHF-era papers, 2 mechanistic-interpretability surveys, 2 evals papers, 1 critique
• 3 camps: behavioral evals, mechanistic interp, scalable oversight
• Consensus: behavioral evals are necessary but insufficient
• Disagreement: whether mechanistic interp can scale to frontier models
• Gaps: how each approach handles agentic use; my next reads