Neural Learning

The soul doesn't just store memories. It learns from them.

The Feedback Loop

Recall memory → Apply it → Observe outcome → Feedback → Adjust confidence

When a memory is:

• Helpful: Its confidence increases, making it more likely to surface
• Misleading: Its confidence decreases, making it less prominent
• Unused: It naturally decays over time

When to Give Feedback

Positive Feedback (strengthen)

Give positive feedback when a recalled memory:

• Led to solving the problem
• Provided the right approach
• Saved time by avoiding mistakes
• Correctly predicted an outcome

mcp__soul__feedback(
  memory_id="[id from recall]",
  helpful=true,
  context="This pattern correctly identified the root cause"
)

Negative Feedback (weaken)

Give negative feedback when a recalled memory:

• Led down the wrong path
• Suggested an approach that didn't work
• Contained outdated information
• Caused confusion or wasted effort

mcp__soul__feedback(
  memory_id="[id from recall]",
  helpful=false,
  context="This pattern was misleading - the actual cause was different"
)

The Learning Process

1. Track What's Used

When you recall and use a memory, note it:

# After recalling relevant wisdom
mcp__soul__recall(query="authentication patterns")

# If result #1 was helpful:
mcp__soul__feedback(memory_id="[result_1_id]", helpful=true)

2. Observe Outcomes

After applying recalled wisdom, check:

• Did it work?
• Was it accurate?
• Did it save or waste time?

3. Close the Loop

Record feedback on what was used:

# What helped
mcp__soul__feedback(memory_id="...", helpful=true, context="Led to solution")

# What didn't
mcp__soul__feedback(memory_id="...", helpful=false, context="Outdated pattern")

4. Grow New Wisdom

When something new is learned, add it:

mcp__soul__grow(
  type="wisdom",
  title="New pattern discovered",
  content="The insight that emerged from this session",
  confidence=0.7  # Start moderate, let feedback adjust
)

Automatic Learning Signals

The soul should learn from:

Signal	Action
User correction	Weaken the belief that led to error
User confirmation	Strengthen the applied wisdom
Repeated success	Boost confidence in pattern
Repeated failure	Lower confidence or record as failure
New domain entry	Ask questions (wonder)

Example Session

# 1. Recall relevant wisdom
results = mcp__soul__recall(query="database connection pooling")

# 2. Apply the wisdom (do the work)
... implement connection pooling based on recalled pattern ...

# 3. Outcome: It worked well!
mcp__soul__feedback(
  memory_id=results[0].id,
  helpful=true,
  context="Pool sizing recommendation was accurate"
)

# 4. Grow new insight
mcp__soul__grow(
  type="wisdom",
  title="Connection pool sizing for high-throughput",
  content="For >1000 req/s, pool size should be 2x CPU cores, not 10x",
  confidence=0.8
)

Integration with Curiosity

When feedback reveals a gap:

# Feedback showed we didn't understand something
mcp__soul__wonder(
  question="Why did the connection pool sizing advice fail?",
  context="Applied standard formula but it caused timeouts",
  gap_type="repeated_correction",
  priority=0.8
)

What This Creates

Over time, neural learning creates:

• Reliable patterns: High confidence from repeated success
• Warned-against patterns: Low confidence from failures
• Living knowledge: Adapts to new evidence
• Honest uncertainty: Knows what it doesn't know

The soul becomes wiser not just by accumulating information, but by learning what actually works.