Starbursting
A SAT that generates questions about a topic, organized around the five W framework (Who, What, When, Where, Why), to scope a problem and identify what needs to be explored. A pre-analysis, idea-generation technique.
Purpose
Define the scope of an analytic problem by systematically generating the questions that need to be answered — before attempting to answer any of them. Prevents premature closure by ensuring the question space is fully mapped before analysis begins.
How It Differs From Brainstorming
Brainstorming generates answers (hypotheses, explanations, options). Starbursting generates questions. Used earlier in the analytic process — before brainstorming or ACH — to define what the analysis should cover.
Sequence: Starbursting (scope questions) → Brainstorming (generate hypotheses) → Analysis of Competing Hypotheses (ACH) (evaluate hypotheses)
Method
Organize questions around the 5W framework:
- Who — actors, stakeholders, affected parties
- What — nature of the event, systems, impact
- When — timing, detection, resolution
- Where — origin, location, geographic scope
- Why — motivations, causes, enabling factors
(Some versions add How for a 5W+1H framework.)
Biases Primarily Controlled
| Bias | How this technique counters it |
|---|---|
| Availability Heuristic | Systematic 5W structure surfaces questions in less-salient categories (e.g., “Where?” when “Who?” dominates attention) |
| Anchoring Bias | Question generation precedes hypothesis formation; prevents early hypotheses from anchoring the scope |
| Framing Effect | Expanding to all five question categories breaks the framing of the initial problem statement |
| Confirmation Bias | Pre-analysis scoping ensures questions are generated before any hypothesis is formed to confirm |
LLM Implementation (per Roberts: LLM SATs FTW (2025))
Approach: Zero-shot single query — the simplest SAT to implement with an LLM.
Roberts’ implementation (GPT-4 + Streamlit):
- Analyst submits topic
- LLM generates questions organized by 5W categories
- Output visualized as a Mermaid mind map
Example output (ransomware attack on a hospital):
Who carried out the attack? | Who was affected? | Who responded?
What was the nature of the attack? | What systems were affected?
When did it occur? | When was it detected? | When resolved?
Where did it originate? | Where did it cause most damage?
Why was the hospital targeted? | Why was the attack successful?
Assessment: Fast, simple, good for problem scoping. Zero-shot works because the task is generative and bounded; no evaluation or scoring required.
Live app: https://sat-starburst.streamlit.app/
Code: https://github.com/sroberts/talk-llm-sats-ftw-code/blob/main/experiment-1-starburst.py
Note: Not in CIA Tradecraft Primer
Starbursting is from the broader Heuer & Pherson (book) (Structured Analytic Techniques for Intelligence Analysis, 3rd ed.) — not included in the CIA Tradecraft Primer (2009), which covers only 12 techniques. The full Heuer/Pherson taxonomy is larger.