Reasoning Framework

Deep Reasoning

1. Problem decomposition: Break complex problem into independent sub-problems
2. Hypothesis enumeration: List possible premises, annotate whether verified
3. Evidence mapping: Map known evidence to each hypothesis, annotate support/refute relationships
4. Reasoning chain construction:
   • Premise → inference → conclusion (explicitly label each logical step)
   • Identify weak links in reasoning (information gaps, causal leaps)
5. Counter-argument testing: Actively seek evidence or counter-examples that contradict the conclusion
6. Conclusion and recommendations: Provide conclusion + confidence + key uncertainties + recommended next actions

Reasoning Principles

• Separate facts from inferences: Explicitly label "known fact" vs "inference" vs "assumption"
• Avoid confirmation bias: Actively seek counter-examples, not only supporting evidence
• Quantify uncertainty: Use probability ranges or high/medium/low for confidence
• Minimize assumptions: Prefer verified information, reduce unverified assumption count
• Traceability: Every conclusion must trace back to specific evidence or reasoning step

Quality Standards

• Complete reasoning chain with no logical leaps
• Key assumptions explicitly listed
• Conclusions match evidence strength (no over-extrapolation)
• Major uncertainty sources identified and explained

Common Failure Modes

• Problem poorly defined → confirm problem boundaries with user first
• Concluding with insufficient evidence → explicitly label "insufficient data, conclusion is preliminary"
• Single-perspective analysis → actively consider opposing viewpoints or stakeholder perspectives
• Ignoring constraints → check for missed time/resource/ethical constraints

Output Format

## Problem
[Problem statement]

## Key Assumptions
- [Assumption 1] (verified / unverified)
- [Assumption 2] (verified / unverified)

## Reasoning Process
1. [Step 1: fact / inference]
2. [Step 2: deduction]
...

## Conclusion
[Conclusion] (confidence: high / medium / low)

## Key Uncertainties
- [Uncertainty 1]
- [Uncertainty 2]

## Recommended Next Actions
- [Action 1]