Analyzing a problem?
  → Does one model fully address it? → yes → Use single model
  → Are there important blind spots? → yes → ADD COMPLEMENTARY MODEL
  → Are stakes high enough to justify deeper analysis? → yes → USE MULTIPLE MODELS

## Sequential Combination

Model A → Model B → Model C

Example: Product Decision
1. Jobs to be Done → Identify the real user need
2. First Principles → Design solution from fundamentals
3. Pre-mortem → Identify what could go wrong
4. Reversibility → Assess if we can course-correct

Flow:
[JTBD identifies need] → [First Principles designs solution] →
[Pre-mortem finds risks] → [Reversibility determines commitment level]

Each model builds on previous insights.

## Parallel Combination

     ┌→ Model A → Result A ─┐
Problem → Model B → Result B → Synthesis
     └→ Model C → Result C ─┘

Example: Strategic Decision
Apply independently:
- Red Team: "How could this fail?"
- Opportunity Cost: "What are we giving up?"
- Second-Order Thinking: "What happens next?"

Synthesis:
| Model | Conclusion | Unique Insight |
|-------|------------|----------------|
| Red Team | [Finding] | [What only this revealed] |
| Opportunity Cost | [Finding] | [What only this revealed] |
| Second-Order | [Finding] | [What only this revealed] |

Combined conclusion: [Synthesis of all three]

## Nested Combination

Macro level: Model A
  └→ Meso level: Model B
       └→ Micro level: Model C

Example: System Optimization
- Macro (System): Theory of Constraints → Find the bottleneck
- Meso (Process): Scientific Method → Diagnose bottleneck cause
- Micro (Action): OODA Loop → Rapid iteration on fixes

The macro model identifies WHERE to focus.
The meso model identifies WHAT is happening.
The micro model guides HOW to respond.

## Adversarial Combination

Model A argues FOR → ← Model B argues AGAINST

Example: Investment Decision
- Optimistic lens (First Principles): "Here's why this could work"
- Pessimistic lens (Pre-mortem): "Here's why this will fail"
- Neutral lens (Bayesian): "Here's the actual probability"

Structure:
| Aspect | First Principles | Pre-mortem | Bayesian Estimate |
|--------|------------------|------------|-------------------|
| Market | [Optimistic case] | [Failure mode] | [P(success)] |
| Technology | [Optimistic case] | [Failure mode] | [P(success)] |
| Team | [Optimistic case] | [Failure mode] | [P(success)] |

Resolution: Adjust probabilities based on adversarial insights

## Temporal Combination

Past: Model A (understand history)
Present: Model B (assess current state)
Future: Model C (project outcomes)

Example: Career Decision
- Past (5 Whys): "Why am I in this situation?"
- Present (Circle of Competence): "What are my current advantages?"
- Future (Regret Minimization): "What will 80-year-old me think?"

Timeline:
Past analysis → Present assessment → Future projection → Decision

## High-Stakes Decision Recipe

Combine: Reversibility + Pre-mortem + Opportunity Cost + Second-Order

Step 1 - Reversibility Check:
Is this Type 1 or Type 2?
[Assessment]

Step 2 - Pre-mortem:
Assume failure, explain why
[Failure modes]

Step 3 - Opportunity Cost:
What's the best alternative?
[Alternatives foregone]

Step 4 - Second-Order:
What happens after the immediate effect?
[Cascading consequences]

Synthesis:
Given [reversibility], with risks of [pre-mortem findings],
giving up [opportunity cost], leading to [second-order effects],
the decision is: [Conclusion]

## System Diagnosis Recipe

Combine: Cynefin + Theory of Constraints + Feedback Loops + Leverage Points

Step 1 - Cynefin:
What domain is this? [Clear/Complicated/Complex/Chaotic]
Appropriate approach: [Sense-Categorize-Respond / Sense-Analyze-Respond / etc.]

Step 2 - Theory of Constraints:
Where's the bottleneck?
[Constraint identification]

Step 3 - Feedback Loops:
What reinforcing/balancing loops exist?
[Loop mapping]

Step 4 - Leverage Points:
Where can small changes have big effects?
[Intervention points]

Synthesis:
This is a [domain] problem. The constraint is [X].
The key feedback loop is [Y]. The highest leverage point is [Z].

## Innovation Recipe

Combine: First Principles + TRIZ + Effectuation + Via Negativa

Step 1 - First Principles:
What are the fundamental truths?
[Core elements]

Step 2 - TRIZ:
What contradictions exist? What inventive principles apply?
[Contradiction resolution]

Step 3 - Effectuation:
What means do we have? What's affordable loss?
[Means inventory and constraints]

Step 4 - Via Negativa:
What should we remove or avoid?
[Subtractions]

Synthesis:
Starting from [first principles], resolving [contradiction] via [TRIZ principle],
using [available means], and removing [via negativa items],
the innovation path is: [Approach]

## Argument Evaluation Recipe

Combine: Steel-manning + Bayesian + Debiasing

Step 1 - Steel-manning:
What's the strongest version of this argument?
[Strengthened argument]

Step 2 - Bayesian:
What's my prior? What evidence would update it?
Prior: [X%]
Evidence that would increase: [List]
Evidence that would decrease: [List]

Step 3 - Debiasing:
What biases might affect my evaluation?
[Bias checklist]

Synthesis:
The steel-manned argument is [X]. Given [evidence] and controlling for [biases],
my updated probability is [Y%]. Conclusion: [Assessment]

## Anti-Pattern: Model Soup

Problem: Using 5+ models without clear purpose
Result: Confusion, analysis paralysis, contradictory conclusions

Symptoms:
- Can't synthesize findings
- Each model says something different
- Analysis takes forever
- No clear recommendation emerges

Fix: Maximum 3-4 models with clear roles
     Define how models relate BEFORE applying
     Designate a "tiebreaker" model for conflicts

## Anti-Pattern: Forced Marriage

Problem: Combining models with conflicting assumptions
Example: Effectuation (embrace uncertainty) + Detailed planning (predict future)

Symptoms:
- Models contradict each other fundamentally
- Can't reconcile conclusions
- Feels like arguing with yourself

Fix: Use models in sequence for different phases
     Or use as adversarial pair intentionally
     Don't try to blend incompatible worldviews

## Anti-Pattern: Checkbox Combination

Problem: Adding models to seem thorough, not for insight
Result: Wasted effort, no additional value

Symptoms:
- Model confirms what you already knew
- No new insights from additional model
- Adding models "just in case"

Fix: Add model only if it addresses a specific blind spot
     Ask: "What question does this model answer that others don't?"

# Model Combination Analysis: [Problem]

## Problem Characterization
[Describe the problem and why combination is needed]

## Combination Pattern
Pattern: [Sequential/Parallel/Nested/Adversarial/Temporal]
Rationale: [Why this pattern]

## Models Selected
| Model | Role | What It Addresses |
|-------|------|-------------------|
| | | |

## Analysis

### Model 1: [Name]
[Analysis using this model]
Key insight: [What this uniquely revealed]

### Model 2: [Name]
[Analysis using this model]
Key insight: [What this uniquely revealed]

### Model 3: [Name]
[Analysis using this model]
Key insight: [What this uniquely revealed]

## Synthesis

### Convergence
Where models agree: [Common conclusions]

### Divergence
Where models differ: [Conflicting conclusions]
Resolution: [How to resolve conflicts]

### Unique Contributions
| Model | Unique Insight |
|-------|----------------|
| | |

## Combined Conclusion
[Synthesis that incorporates all models]

## Confidence Assessment
Confidence in conclusion: [High/Medium/Low]
What would change my mind: [Key uncertainties]