Single-Loop Learning

Single-Loop Learning

Correct action within existing frame.

When to Use

• Action produced unexpected result
• Fix needed without questioning goals
• First occurrence (or second) of problem
• Frame/goals have been validated

Mental Model

Thermostat: Detect deviation from set point → adjust action → return to target.

The set point (goal) is not questioned. Only the action is adjusted.

Goal (fixed) → Action → Result → Gap detected → Adjust action → Repeat

Single-Loop vs Double-Loop

Process

1. Detect Gap

Identify the deviation between expected and actual.

Questions:

• What did we expect to happen?
• What actually happened?
• How large is the gap?
• How urgent is the fix?

2. Diagnose Cause

Find the immediate cause (not root cause — that's double-loop territory).

Techniques:

• 5 Whys (shallow): Stop at proximate cause, not systemic
• Fault tree: Map possible causes, identify which occurred
• Timeline: What changed right before the problem?

Key distinction:

• Proximate cause: "Query filter was wrong"
• Root cause: "No validation in process" (save for prevention step)

3. Identify Correction

Determine what action to change.

Correction types:

• Parameter adjustment: Change a value (threshold, limit, timeout)
• Process addition: Add a step (validation, check, review)
• Input validation: Verify inputs before processing
• Error handling: Add fallback when failure occurs
• Monitoring: Detect problem faster next time

4. Verify Fix

Confirm the correction works.

Questions:

• Does the corrected action produce expected result?
• Any side effects?
• Is the fix complete?

5. Implement Prevention

Reduce likelihood of recurrence.

Prevention types:

• Add check/validation step
• Add monitoring/alerting
• Update documentation
• Add to checklist

Frame Validity Check

Before completing single-loop, verify the frame is still valid:

Check:

• Is this the first time this type of failure occurred? (1-2 = single-loop OK)
• Did the corrected action achieve the goal? (Yes = frame OK)
• Are underlying assumptions still valid? (Yes = frame OK)

Escalate to double-loop if:

• Third failure of same type
• Correction doesn't achieve goal
• Assumptions seem wrong

Output Format

## Single-Loop: [Problem]

**Gap detected:**
- Expected: [What should have happened]
- Actual: [What happened]
- Severity: [Critical/High/Medium/Low]

**Diagnosis:**
- Immediate cause: [What went wrong]
- Contributing factors: [What enabled it]
- Investigation: [How we found it]

**Correction:**
- From: [Old action/value]
- To: [New action/value]
- Rationale: [Why this fixes it]

**Verification:**
- Test: [How we verified]
- Result: [Pass/Fail]

**Prevention:**
- Immediate: [What we added to prevent recurrence]
- Process change: [Any process updates]
- Monitoring: [Any new monitoring]

**Frame check:**
- Pattern? [Yes/No — if Yes, escalate to double-loop]
- Frame valid? [Yes/No]

Common Patterns

Parameter Was Wrong

Diagnosis: Threshold too low/high
Correction: Adjust parameter
Prevention: Document correct range

Step Was Missing

Diagnosis: No validation before action
Correction: Add validation step
Prevention: Add to checklist

Input Was Bad

Diagnosis: Garbage in, garbage out
Correction: Add input validation
Prevention: Validate at boundary

Error Not Handled

Diagnosis: Failure case not covered
Correction: Add error handling
Prevention: Review error paths

Anti-Patterns