| name | Erotetic |
| description | Question-driven reasoning based on Koralus's Erotetic Theory. USE WHEN erotetic, question-driven reasoning, enumerate alternatives, check reasoning, reasoning audit, fallacy check, what am I not considering, stable conclusion. |
Erotetic Reasoning Skill
Question-driven reasoning grounded in Philipp Koralus's Erotetic Theory of Reasoning (Oxford, 2023). Forces explicit alternative enumeration, systematic question-raising, and equilibrium checking before settling on conclusions.
Core insight: The same mechanism that produces valid reasoning also produces fallacies — the variable is how many questions you raise before settling. More questions → closer to erotetic equilibrium → convergence with classical validity, probabilistic coherence, and expected utility maximization.
How it differs from other thinking approaches:
- First Principles deconstructs constraints and rebuilds from fundamentals
- Devil's Advocate / Red Team attacks arguments from adversarial perspectives
- Chain-of-thought asks to "think step by step" without a formal stopping criterion
- Erotetic forces you to enumerate ALL alternatives, raise the questions you haven't asked, and test whether conclusions survive further questioning
Workflow Routing
| Workflow | Trigger | File |
|---|
| Reason | "think through this erotetically", "what am I not considering", "enumerate alternatives", "question-driven analysis" | Workflows/Reason.md |
| Audit | "audit this reasoning", "check for fallacies", "is this conclusion stable", "erotetic audit" | Workflows/Audit.md |
Quick Reference
The Erotetic Reasoning Loop:
- Enumerate — List ALL disjunctive alternatives (not just the obvious ones)
- Question — What questions does this raise? What haven't I asked?
- Filter — Given new information, which alternatives survive?
- Equilibrium Check — Is this conclusion stable under one more question?
- Iterate or Settle — If unstable, return to step 2. If stable, conclude.
ETR-Predicted Fallacies (what to watch for):
- Settling on first best-match without enumerating alternatives (conjunction fallacy)
- Ignoring base rates when a vivid alternative is present
- Illusory inferences from disjunctions
- Premature closure — stopping questioning too early
Examples
Example 1: Strategic decision
User: "Should I raise prices 20%? Think through this erotetically."
→ Invokes Reason workflow
→ Enumerates: price stays / +10% / +20% / +30% / restructure pricing / freemium shift
→ Raises questions: What's the churn elasticity? What do competitors charge?
→ Filters alternatives based on available evidence
→ Checks equilibrium: "Is +20% stable if I ask one more question?"
→ Returns structured analysis with equilibrium assessment
Example 2: Audit existing reasoning
User: "I concluded we should use microservices. Audit this reasoning."
→ Invokes Audit workflow
→ Identifies: What alternatives were considered? (monolith, modular monolith, serverless?)
→ Checks: Was this a best-match latch without enumeration?
→ Tests equilibrium: Does "microservices" survive the question "what's our team size?"
→ Returns fallacy classification and stability assessment
Example 3: Combined with other skills
User: "Red team this idea, but use erotetic reasoning"
→ Invoke adversarial analysis with Erotetic as the reasoning framework
→ Each agent enumerates alternatives and raises questions
→ Equilibrium check on the steelman and counter-argument
Integration Pattern
Other skills can invoke Erotetic reasoning:
## Before Concluding
→ Use Erotetic/Reason to enumerate alternatives you haven't considered
→ Use Erotetic/Audit to check if your conclusion is stable
## When Stuck
→ Use Erotetic/Reason — the bottleneck is usually an unasked question
## For High-Stakes Decisions
→ Adversarial analysis + Erotetic: attack with explicit alternative enumeration
→ First Principles + Erotetic: deconstruct, then question every reconstruction
Theoretical Foundation
- Source: Koralus, P. (2023). Reason and Inquiry. Oxford University Press.
- Formalization: PyETR (Oxford HAI Lab)
- Key result: At erotetic equilibrium, reasoning converges to classical validity, probabilistic coherence, AND expected utility maximization
- AI finding: Richardson et al. (2025) — as LLMs scale, errors become more human-like in ETR-predicted ways; scaling alone does not solve reasoning
