/compound

Extract and save learnings from completed tasks to build compound organizational intelligence.

When to Use

• After completing a significant task or deliverable
• After a project milestone or sprint
• When something went wrong (or unexpectedly right)
• Periodically as a reflection habit (weekly/monthly)

Vision to Value Phase

Phase 6: Learning & Adaptation (but usable after any phase)

Workflow

Step 1: Extract

Reflect on the completed work:

Question	Purpose
What was the task?	Clear scope
What was the outcome?	Actual result
What worked well?	Repeat these
What didn't work?	Avoid these
What surprised us?	New insights
What took longer than expected?	Estimation calibration

Step 2: Root Cause

For things that didn't work or surprised:

• Why did this happen? (5 Whys if needed)
• Was it a process issue, knowledge gap, tool problem, or assumption error?
• Was it preventable with information we had at the time?

Step 3: Generalize

Turn specific observations into reusable principles:

Specific	Generalized Learning
"The API changed and broke our integration"	"External API dependencies need health monitoring"
"The customer interview revealed a need we hadn't considered"	"Always include open-ended questions in discovery interviews"
"Estimating this feature took 3x longer than planned"	"Features touching 3+ system boundaries need 2x estimation buffer"

Step 4: Save

Save each learning to

context/learnings/

using

/context-save

:

# L-NNN: [Learning Title]

**Date**: [YYYY-MM-DD]
**Source**: [Task/project that generated this]
**Category**: [Process | Technical | Customer | Market | Team | Tool]
**Confidence**: [High | Medium | Low] (proven once vs multiple times)

## Learning
[1-2 sentence generalized principle]

## Evidence
[What specifically happened that taught us this]

## Application
[When should someone recall this learning]

## Related
[Links to decisions, bets, or other learnings]

Step 5: Index

Update the learnings index so future

/context-recall

and

/relevant-learnings

can find them:

• Add to

  context/learnings/index.md

• Update topic tags in

  context/index.json

• Cross-reference to related decisions/bets if applicable

Triggers

Manual (user-initiated)

• "Let's do a compound learning extraction"
• "What did we learn from [project]?"
• After

  /retrospective

  or

  /outcome-review

Suggested (agent-initiated)

After completing significant work, agents should offer:

"Want me to extract learnings from this work? (

/compound

)"

Periodic

• End of week: Quick scan of the week's work
• End of project phase: Deeper extraction
• After failures or surprises: Immediate extraction

Output Format

# Compound Learning: [Context/Project]

**Date**: [YYYY-MM-DD]
**Scope**: [What work this covers]

## Learnings Extracted

### L-NNN: [Title]
**Category**: [category]
**Learning**: [principle]
**Evidence**: [what happened]
**Application**: [when to recall]

### L-NNN+1: [Title]
...

## Updated Indexes
- Added [N] learnings to `context/learnings/index.md`
- Cross-referenced to: [list of related IDs]

## Prevention Measures
- [Concrete changes to prevent recurring issues]
- [Process adjustments recommended]

Integration

• Feeds from:

  /retrospective

  ,

  /outcome-review

  ,

  /decision-quality-audit

• Outputs to:

  context/learnings/

  ,

  context/index.json

• Retrieved by:

  /context-recall

  ,

  /relevant-learnings

• Cross-references: decisions, bets, assumptions

Based On

• Compound learning principle: each learning makes the next task better
• Double-loop learning (Argyris): not just fixing errors but questioning the reasoning
• Blameless post-mortems (SRE culture)