Decision Review

Purpose: Analyze how a past decision played out to improve future decision-making.

Announce: "Using decision-review to analyze how this decision played out."

When to Use

Invoke this skill when:

•A scheduled review date has arrived
•An exit trigger has activated
•Sufficient time has passed to observe outcomes
•Circumstances have materially changed
•You want to learn from a past decision (success or failure)
•Building a track record for calibration purposes

Process

Step 1: Retrieve Original Decision

Load the decision artifact and review what was decided.

Questions to answer:

•What was the decision?
•What was the recommendation?
•What were the Must-Be-True conditions?
•What risks were identified?
•What Exit Criteria were established?
•What alternatives were rejected?

markdown

## Original Decision Summary

**Decision Date:** [YYYY-MM-DD]
**Decision:** [What was decided]
**Recommendation at time:** [Proceed/Do Not Proceed]
**Confidence at time:** [High/Medium/Low]

**Must-Be-True Conditions:**
1. [Condition 1]
2. [Condition 2]
3. [Condition 3]

**Identified Risks:**
1. [Risk 1]
2. [Risk 2]
3. [Risk 3]

**Exit Criteria:**
- [Tripwire 1]
- [Tripwire 2]

Step 2: Document Outcomes

Record what actually happened since the decision was made.

Outcome Assessment Questions:

•What was the actual result?
•How does it compare to expectations?
•What unexpected events occurred?
•Which exit triggers, if any, were hit?
•What's the current status?

Outcome Categories:

Category	Description
Success	Objectives achieved, decision validated
Partial Success	Some objectives achieved, some missed
Neutral	Neither clear success nor failure
Partial Failure	Significant objectives missed
Failure	Decision did not achieve intended outcomes
Reversed	Decision was undone or significantly changed

markdown

## Actual Outcomes

**Review Date:** [YYYY-MM-DD]
**Time Since Decision:** [duration]
**Overall Outcome:** [Success/Partial Success/Neutral/Partial Failure/Failure/Reversed]

**What Happened:**
[Narrative description of actual events and outcomes]

**Unexpected Events:**
- [Event 1]
- [Event 2]

**Exit Triggers Hit:**
- [ ] [Trigger 1] - [hit/not hit]
- [ ] [Trigger 2] - [hit/not hit]

**Current Status:**
[Where things stand now]

Step 3: Analyze Predictions

Compare predictions against reality to calibrate future judgment.

Must-Be-True Conditions Accuracy:

For each condition identified at decision time:

•Did it hold true?
•If not, what happened instead?
•How critical was the failure?

Risk Predictions Accuracy:

For each risk identified:

•Did it materialize?
•Was severity/likelihood accurate?
•Were mitigations effective?

Unidentified Factors:

•What important factors were not anticipated?
•What risks materialized that weren't on the list?
•What opportunities emerged unexpectedly?

markdown

## Prediction Analysis

### Must-Be-True Conditions

| Condition | Held True? | What Actually Happened |
|-----------|------------|----------------------|
| [Condition 1] | Yes/No/Partial | [Reality] |
| [Condition 2] | Yes/No/Partial | [Reality] |
| [Condition 3] | Yes/No/Partial | [Reality] |

**Condition Accuracy:** [X of Y conditions held true]

### Risk Predictions

| Risk | Materialized? | Predicted L/I | Actual L/I | Mitigation Worked? |
|------|---------------|---------------|------------|-------------------|
| [Risk 1] | Yes/No | [H/M/L] / [H/M/L] | [H/M/L] / [H/M/L] | Yes/No/Partial |
| [Risk 2] | Yes/No | [H/M/L] / [H/M/L] | [H/M/L] / [H/M/L] | Yes/No/Partial |
| [Risk 3] | Yes/No | [H/M/L] / [H/M/L] | [H/M/L] / [H/M/L] | Yes/No/Partial |

**Risk Prediction Accuracy:** [X of Y risks correctly assessed]

### Blind Spots

Factors we didn't anticipate:
1. [Blind spot 1 - what was missed]
2. [Blind spot 2 - what was missed]

Step 4: Identify Lessons

Extract actionable learning from the analysis.

What We Got Right:

•Which predictions were accurate?
•Which parts of the process worked well?
•What should we repeat?

What We Got Wrong:

•Which predictions failed?
•Where did analysis fall short?
•What should we avoid?

What We Should Do Differently:

•Process improvements for future decisions
•New questions to ask
•New checks to perform
•Biases to watch for

markdown

## Key Lessons

### What We Got Right
- [Lesson 1]
- [Lesson 2]

### What We Got Wrong
- [Lesson 1]
- [Lesson 2]

### Process Improvements
For future similar decisions:
1. [Improvement 1]
2. [Improvement 2]
3. [Improvement 3]

### Bias Observations
Biases that affected this decision:
- [Bias 1 and how it manifested]
- [Bias 2 and how it manifested]

Step 5: Update Patterns

Document insights for use in future decisions.

Pattern Recognition:

•Does this decision fit a pattern of past successes or failures?
•What decision type does this represent?
•What heuristics should be updated?

Calibration Update:

•How should this affect confidence in similar future decisions?
•Were we overconfident or underconfident?
•What's our track record in this domain now?

Knowledge Base Updates:

•New facts to add to institutional knowledge
•Updated risk assessments for similar situations
•Revised assumptions about key variables

markdown

## Pattern Updates

### Decision Type
[Category of decision - e.g., vendor selection, strategic pivot, hiring]

### Updated Heuristics
For similar decisions in the future:
- [Heuristic 1]
- [Heuristic 2]

### Calibration Adjustment
- Previous confidence level: [X]
- Recommended adjustment: [raise/lower/maintain]
- Track record in this domain: [X successes / Y failures]

### Institutional Knowledge
Add to knowledge base:
- [Fact/insight 1]
- [Fact/insight 2]

Output Template

Generate a comprehensive review document:

markdown

# Decision Review: [Decision Name]

## Summary

| Aspect | Original | Actual |
|--------|----------|--------|
| **Decision Date** | [YYYY-MM-DD] | - |
| **Review Date** | - | [YYYY-MM-DD] |
| **Time Elapsed** | - | [duration] |
| **Recommendation** | [Proceed/Do Not Proceed] | - |
| **Confidence** | [High/Medium/Low] | - |
| **Outcome** | - | [Success/Partial/Failure] |

## Prediction Accuracy

### Must-Be-True Conditions
**Accuracy:** [X/Y] conditions held true

| Condition | Predicted | Actual | Match |
|-----------|-----------|--------|-------|
| [Condition 1] | Must hold | [Yes/No] | [check/X] |
| [Condition 2] | Must hold | [Yes/No] | [check/X] |

### Risk Predictions
**Accuracy:** [X/Y] risks correctly predicted

| Risk | Predicted | Materialized | Severity Match |
|------|-----------|--------------|----------------|
| [Risk 1] | [L/I] | [Yes/No] | [check/X] |
| [Risk 2] | [L/I] | [Yes/No] | [check/X] |

### Blind Spots
- [Unanticipated factor 1]
- [Unanticipated factor 2]

## Key Lessons

### What Worked
1. [Success factor 1]
2. [Success factor 2]

### What Didn't Work
1. [Failure factor 1]
2. [Failure factor 2]

### Process Improvements
1. [Improvement 1]
2. [Improvement 2]

## Recommendations

### For Similar Future Decisions
- [Recommendation 1]
- [Recommendation 2]

### Calibration Update
[How this affects confidence in similar decisions]

### Follow-up Actions
- [ ] [Action 1]
- [ ] [Action 2]

---

*Review conducted: [date]*
*Reviewer: [name/role]*
*Original decision: [link to decision artifact]*

Save Location

Save the review document to:

code

docs/decisions/YYYY-MM-DD-<decision-slug>/review-[date].md

Where:

•YYYY-MM-DD-<decision-slug> matches the original decision directory
•[date] is the review date in YYYY-MM-DD format

Example: docs/decisions/2024-03-15-vendor-selection/review-2024-09-15.md

Exit Criteria

•Original decision artifact retrieved and summarized
•Actual outcomes documented with evidence
•All Must-Be-True conditions assessed
•All predicted risks evaluated
•Blind spots identified
•Lessons extracted and actionable
•Patterns documented for future use
•Review document saved to correct location

Integration with Calibration

Reviews feed the calibration system:

•Track prediction accuracy over time
•Identify systematic biases in forecasting
•Build domain-specific confidence adjustments
•Improve the decision-making process itself

After completing a review, consider:

•Is this a pattern that affects other pending decisions?
•Should exit criteria on other decisions be updated?
•Does this change institutional risk tolerance?