SAT Selection Guide
Query: Given a problem type, bias risk, or point in the analytic process — which SAT(s) should I apply?
Use this page to select the right technique. For how techniques work together in sequence, see SAT Pipeline. For the full cross-reference, see Bias x SAT Matrix.
Select by: Where Are You in the Process?
| Stage | Problem | Reach For |
|---|---|---|
| Before you start | Need to scope what questions to ask | Starbursting |
| Before you start | Need to generate candidate explanations | Brainstorming |
| During analysis | Have a preferred explanation and want to test it rigorously | ACH |
| During analysis | Need to surface hidden assumptions in the current line | Key Assumptions Check |
| During analysis | Need to evaluate source reliability | Quality of Information Check |
| During analysis | Need an adversary’s perspective | Red Team Analysis |
| During analysis | Need to monitor for changing conditions | Indicators or Signposts of Change |
| Challenging a conclusion | A consensus has formed; want to stress-test it | Devil’s Advocacy |
| Challenging a conclusion | Need truly independent views (not just devil’s advocate) | Team B |
| Planning / forecasting | Need to think about uncertain futures | Alternative Futures Analysis |
| Planning / forecasting | Need to consider low-probability but high-impact events | Low-Probability Analysis |
| Post-analysis | Need to stress-test a finished product | What If? Analysis, Key Assumptions Check |
Select by: Which Bias Are You Worried About?
Select by: Problem Type
Attribution / “Who did this?”
Primary risk: Confirmation Bias, Mirror Imaging, Anchoring Bias
Recommended sequence:
- Brainstorming — generate all possible actors
- ACH — evaluate evidence against all actors simultaneously
- Red Team Analysis — would the attributed actor actually behave this way?
Forecasting / “What will happen?”
Primary risk: Status Quo Bias, Overconfidence Bias, Availability Heuristic
Recommended sequence:
- Alternative Futures Analysis — develop multiple scenarios from key uncertainties
- Low-Probability Analysis — explicitly address the tail
- Indicators or Signposts of Change — define observable signals for each scenario
Reviewing a finished product / “Is this analysis sound?”
Primary risk: Confirmation Bias, Anchoring Bias, Hallucination (LLM)
Recommended sequence:
- Key Assumptions Check — surface all load-bearing assumptions
- Devil’s Advocacy — build the best case against the conclusion
- Quality of Information Check — audit sources and evidence quality
Planning / “What could go wrong?”
Primary risk: Overconfidence Bias, Status Quo Bias, Motivated Reasoning
Recommended sequence:
- What If? Analysis — assume it has already failed; work backwards
- Low-Probability Analysis — enumerate the tail risks
- Key Assumptions Check — identify which assumptions, if wrong, cause the plan to fail
Adversarial / “What will an opponent do?”
Primary risk: Mirror Imaging, Framing Effect, Overconfidence Bias
Recommended sequence:
- Outside-In Thinking — force analysis from the adversary’s external constraints
- Red Team Analysis — adopt adversary persona and evaluate the plan from their perspective
- What If? Analysis — assume the adversary succeeded; reconstruct how
Multi-agent deliberation / “How do we prevent echo chambers?”
Primary risk: Groupthink, Sycophancy, Confirmation Bias
Recommended pattern:
- Run agents independently before any agent sees others’ outputs (Team B pattern)
- Assign one agent an explicit adversarial critique role (Devil’s Advocacy pattern)
- Use different system prompts or models to prevent synchronization
- See SAT Pipeline for the full multi-agent architecture
Minimum Viable Intervention
When you can only apply one SAT and bias risk is general, these deliver the broadest coverage:
| Priority | SAT | Biases covered | LLM implementation complexity |
|---|---|---|---|
| 1st | Key Assumptions Check | Anchoring, Confirmation, Overconfidence, Motivated Reasoning, Status Quo | Single prompt; zero-shot |
| 2nd | Devil’s Advocacy | Confirmation, Groupthink, Motivated Reasoning, Sycophancy | Single adversarial prompt; or two-agent |
| 3rd | ACH | Confirmation, Anchoring, Availability, Motivated Reasoning, Groupthink | Multi-step sequential (see Roberts: LLM SATs FTW (2025)) |