forecast-premortem

What this skill does

# Forecast Pre-Mortem

## Table of Contents
- [Interactive Menu](#interactive-menu)
- [Quick Reference](#quick-reference)
- [Resource Files](#resource-files)

**Core Principle**: Invert the problem. Instead of "Will this succeed?", ask "It has failed -- why?"

---

## Interactive Menu

**What would you like to do?**

### Core Workflows

**1. [Run a Failure Premortem](#1-run-a-failure-premortem)** - Assume prediction failed, explain why
**2. [Run a Success Premortem](#2-run-a-success-premortem)** - For pessimistic predictions (<20%)
**3. [Dragonfly Eye Perspective](#3-dragonfly-eye-perspective)** - View failure through multiple lenses
**4. [Identify Tail Risks](#4-identify-tail-risks)** - Find black swans and unknown unknowns
**5. [Adjust Confidence Intervals](#5-adjust-confidence-intervals)** - Quantify the adjustment
**6. [Learn the Framework](#6-learn-the-framework)** - Deep dive into methodology
**7. Exit** - Return to main forecasting workflow

---

## 1. Run a Failure Premortem

**Let's stress-test your prediction by imagining it has failed.**

```
Failure Premortem Progress:
- [ ] Step 1: State the prediction and current confidence
- [ ] Step 2: Time travel to failure
- [ ] Step 3: Write the history of failure
- [ ] Step 4: Identify concrete failure modes
- [ ] Step 5: Assess plausibility and adjust
```

### Step 1: State the prediction and current confidence

**Tell me:**
1. What are you predicting?
2. What's your current probability?
3. What's your confidence interval?

**Example:** "This startup will reach $10M ARR within 2 years" - Probability: 75%, CI: 60-85%

### Step 2: Time travel to failure

**The Crystal Ball Exercise:**

Jump forward to the resolution date. **It is now [resolution date]. The event did NOT happen.** This is a certainty. Do not argue with it.

**How does it feel?** Surprising? Expected? Shocking? This emotional response tells you about your true confidence.

### Step 3: Write the history of failure

**Backcasting Narrative:** Starting from the failure point, work backward in time. Write the story of how we got here.

**Prompts:**
- "The headlines that led to this were..."
- "The first sign of trouble was when..."
- "In retrospect, we should have known because..."
- "The critical mistake was..."

**Frameworks to consider:**
- **Internal friction:** Team burned out, co-founders fought, execution failed
- **External shocks:** Regulation changed, competitor launched, market shifted
- **Structural flaws:** Unit economics didn't work, market too small, tech didn't scale
- **Black swans:** Pandemic, war, financial crisis, unexpected disruption

See [Failure Mode Taxonomy](resources/failure-mode-taxonomy.md) for comprehensive categories.

### Step 4: Identify concrete failure modes

**Extract specific, actionable failure causes from your narrative.**

For each failure mode: (1) What happened, (2) Why it caused failure, (3) How likely it is, (4) Early warning signals

**Example:**
| Failure Mode | Mechanism | Likelihood | Warning Signals |
|--------------|-----------|------------|-----------------|
| Key engineer quit | Lost technical leadership, delayed product | 15% | Declining code commits, complaints |
| Competitor launched free tier | Destroyed unit economics | 20% | Hiring spree, beta leaks |
| Regulation passed | Made business model illegal | 5% | Proposed legislation, lobbying |

### Step 5: Assess plausibility and adjust

**The Plausibility Test:**

Ask yourself:
- **How easy was it to write the failure narrative?**
  - Very easy → Drop confidence by 15-30%
  - Very hard, felt absurd → Confidence was appropriate
- **How many plausible failure modes did you identify?**
  - 5+ modes each >5% likely → Too much uncertainty for high confidence
  - 1-2 modes, low likelihood → Confidence can stay high
- **Did you discover any "unknown unknowns"?**
  - Yes, multiple → Widen confidence intervals by 20%
  - No, all known risks → Confidence appropriate

**Quantitative Method:** Sum the probabilities of failure modes:
```
P(failure) = P(mode_1) + P(mode_2) + ... + P(mode_n)
```

If this sum is greater than `1 - your_current_probability`, your probability is too high.

**Example:** Current success: 75% (implied failure: 25%), Sum of failure modes: 40%
**Conclusion:** Underestimating failure risk by 15%, **Adjusted:** 60% success

**Next:** Return to [menu](#interactive-menu) or document findings

---

## 2. Run a Success Premortem

**For pessimistic predictions - assume the unlikely success happened.**

```
Success Premortem Progress:
- [ ] Step 1: State pessimistic prediction (<20%)
- [ ] Step 2: Time travel to success
- [ ] Step 3: Write the history of success
- [ ] Step 4: Identify how you could be wrong
- [ ] Step 5: Assess and adjust upward if needed
```

### Step 1: State pessimistic prediction

**Tell me:** (1) What low-probability event are you predicting? (2) Why is your confidence so low?

**Example:** "Fusion energy will be commercialized by 2030" - Probability: 10%, Reasoning: Technical challenges too great

### Step 2: Time travel to success

**It is now 2030. Fusion energy is commercially available.** This happened. It's real. How?

### Step 3: Write the history of success

**Backcasting the unlikely:** What had to happen for this to occur?
- "The breakthrough came when..."
- "We were wrong about [assumption] because..."
- "The key enabler was..."
- "In retrospect, we underestimated..."

### Step 4: Identify how you could be wrong

**Challenge your pessimism:**
- Are you anchoring too heavily on current constraints?
- Are you underestimating exponential progress?
- Are you ignoring parallel approaches?
- Are you biased by past failures?

### Step 5: Assess and adjust upward if needed

If success narrative was surprisingly plausible, increase probability.

**Next:** Return to [menu](#interactive-menu)

---

## 3. Dragonfly Eye Perspective

**View the failure through multiple conflicting perspectives.**

The dragonfly has compound eyes that see from many angles simultaneously. We simulate this by adopting radically different viewpoints.

```
Dragonfly Eye Progress:
- [ ] Step 1: The Skeptic (why this will definitely fail)
- [ ] Step 2: The Fanatic (why failure is impossible)
- [ ] Step 3: The Disinterested Observer (neutral analysis)
- [ ] Step 4: Synthesize perspectives
- [ ] Step 5: Extract robust failure modes
```

### Step 1: The Skeptic

**Channel the harshest critic.** You are a short-seller, a competitor, a pessimist. Why will this DEFINITELY fail?

**Be extreme:** Assume worst case, highlight every flaw, no charity, no benefit of doubt

**Output:** List of failure reasons from skeptical view

### Step 2: The Fanatic

**Channel the strongest believer.** You are the founder's mother, a zealot, an optimist. Why is failure IMPOSSIBLE?

**Be extreme:** Assume best case, highlight every strength, maximum charity and optimism

**Output:** List of success reasons from optimistic view

### Step 3: The Disinterested Observer

**Channel a neutral analyst.** You have no stake in the outcome. You're running a simulation, analyzing data dispassionately.

**Be analytical:** No emotional investment, pure statistical reasoning, reference class thinking

**Output:** Balanced probability estimate with reasoning

### Step 4: Synthesize perspectives

**Find the overlap:** Which failure modes appeared in ALL THREE perspectives?
- Skeptic mentioned it
- Even fanatic couldn't dismiss it
- Observer identified it statistically

**These are your robust failure modes** - the ones most likely to actually happen.

### Step 5: Extract robust failure modes

**The synthesis:**

| Failure Mode | Skeptic | Fanatic | Observer | Robust? |
|--------------|---------|---------|----------|---------|
| Market too small | Definitely | Debatable | Base rate suggests yes | YES |
| Execution risk | Definitely | No way | 50/50 | Maybe |
| Tech won't scale | Definitely | Already solved | Unknown | Investigate |

Focus adjustment on the **robust** failures