Evolution Strategies

Deep dive into EVOLVE_STRATEGY presets and intent balancing

Overview

Evolution strategies control the balance between repair, optimize, and innovate intents. They inform the selector logic and adjust the circuit breaker behavior.

EVOLVE_STRATEGY=balanced|innovate|harden|repair-only|early-stabilize|steady-state|auto

Location: src/gep/strategy.js

---

Strategy Presets

balanced (Default)

Normal operation. Steady growth with stability.

{
  repair: 0.20,
  optimize: 0.30,
  innovate: 0.50,
  repairLoopThreshold: 0.50,
  label: 'Balanced'
}

• 20% repair: Fix critical errors
• 30% optimize: Refactor and improve existing code
• 50% innovate: Add new features
• Circuit breaker: Triggers when >50% of last 8 cycles were failed repairs

When to use:

• Production systems with moderate error rates
• Continuous improvement workflows
• Default for most use cases

---

innovate

System is stable. Maximize new features and capabilities.

{
  repair: 0.05,
  optimize: 0.15,
  innovate: 0.80,
  repairLoopThreshold: 0.30,
  label: 'Innovation Focus'
}

• 5% repair: Minimal firefighting
• 15% optimize: Minor improvements
• 80% innovate: Aggressive feature development
• Circuit breaker: Very sensitive (30% repair ratio triggers innovation)

When to use:

• After a successful hardening phase
• Low error rate environments
• Rapid prototyping

Do not use during incident response or high error rate periods.

---

harden

After a big change. Focus on stability and robustness.

{
  repair: 0.40,
  optimize: 0.40,
  innovate: 0.20,
  repairLoopThreshold: 0.70,
  label: 'Hardening'
}

• 40% repair: Fix bugs introduced by recent changes
• 40% optimize: Improve reliability and performance
• 20% innovate: Limited new features
• Circuit breaker: Tolerates more repair cycles (70% threshold)

When to use:

• After major feature releases
• Post-incident recovery
• Before production deployment

---

repair-only

Emergency. Fix everything before doing anything else.

{
  repair: 0.80,
  optimize: 0.20,
  innovate: 0.00,
  repairLoopThreshold: 1.00,
  label: 'Repair Only'
}

• 80% repair: Focus on fixing errors
• 20% optimize: Refactor broken code
• 0% innovate: NO new features
• Circuit breaker: Never triggers (allows indefinite repair cycles)

When to use:

• Incident response
• Critical production issues
• High error rate environments (>50% of logs are errors)

This mode disables innovation entirely. Use only during emergencies.

---

early-stabilize

First cycles. Prioritize fixing existing issues before innovating.

{
  repair: 0.60,
  optimize: 0.25,
  innovate: 0.15,
  repairLoopThreshold: 0.80,
  label: 'Early Stabilization'
}

• 60% repair: Fix initial bugs
• 25% optimize: Improve existing code
• 15% innovate: Small feature additions
• Circuit breaker: High tolerance (80% threshold)

When to use:

• First 5 evolution cycles (auto-applied)
• New workspace setup
• After major refactoring

Auto-detection: Automatically applied for cycles 1-5 when no explicit strategy is set.

Location: src/gep/strategy.js:100-104

---

steady-state

Evolution saturated. Maintain existing capabilities. Minimal innovation.

{
  repair: 0.60,
  optimize: 0.30,
  innovate: 0.10,
  repairLoopThreshold: 0.90,
  label: 'Steady State'
}

• 60% repair: Keep system healthy
• 30% optimize: Gradual improvements
• 10% innovate: Rare new features
• Circuit breaker: Very high tolerance (90% threshold)

When to use:

• Long-running production systems
• After reaching desired feature set
• Low change environments

Auto-detection: Triggered by evolution_saturation or force_steady_state signals.

Location: src/gep/strategy.js:107-111

---

auto

Adaptive strategy selection based on cycle count and saturation signals.

EVOLVE_STRATEGY=auto

Heuristics:

1. Cycles 1-5: Apply early-stabilize
2. Saturation detected: Apply steady-state
3. Otherwise: Use balanced

Saturation signals:

• force_steady_state
• evolution_saturation

Location: src/gep/strategy.js:85-119

---

Intent Balancing

Strategies inform the selector via intent allocation ratios. The selector scores genes based on category match:

// Example: balanced strategy
if (gene.category === 'repair') score += 0.20;
if (gene.category === 'optimize') score += 0.30;
if (gene.category === 'innovate') score += 0.50;

The highest-scoring gene is selected.

---

Circuit Breaker

The circuit breaker detects repair loops and forces innovation to break the cycle.

How it works

1. Read last N events from events.jsonl (default: 3)
2. Check if all are failed repairs with the same gene
3. If true and streak ≥ threshold: set FORCE_INNOVATION=true
4. Selector switches to innovation intent

Location: src/evolve.js:787-811

Thresholds by strategy

Strategy	Repair Loop Threshold
balanced	0.50 (50%)
innovate	0.30 (30%)
harden	0.70 (70%)
repair-only	1.00 (never triggers)
early-stabilize	0.80 (80%)
steady-state	0.90 (90%)

Example: With balanced strategy, if 4 out of last 8 cycles were failed repairs (50%), the circuit breaker forces innovation.

---

Backward Compatibility

FORCE_INNOVATION

Legacy environment variable. Maps to innovate strategy.

FORCE_INNOVATION=true
# Equivalent to:
EVOLVE_STRATEGY=innovate

Location: src/gep/strategy.js:90-93

---

Usage Examples

Production hardening after release

EVOLVE_STRATEGY=harden node index.js --loop

Emergency incident response

EVOLVE_STRATEGY=repair-only node index.js --loop

Rapid prototyping

EVOLVE_STRATEGY=innovate node index.js --loop

Adaptive long-term operation

EVOLVE_STRATEGY=auto node index.js --loop

---

Custom Strategies

To define custom strategies, edit src/gep/strategy.js:

var STRATEGIES = {
  'my-custom': {
    repair: 0.30,
    optimize: 0.40,
    innovate: 0.30,
    repairLoopThreshold: 0.60,
    label: 'My Custom Strategy',
    description: 'Balanced with higher optimize focus',
  },
};

Then:

EVOLVE_STRATEGY=my-custom node index.js

---

Strategy Selection Flow

---

Best Practices

1. Start with auto: Let the evolver adapt to your workspace lifecycle
2. Monitor circuit breaker: Frequent forced innovation indicates gene quality issues
3. Use repair-only sparingly: Only for emergencies
4. Harden after innovation sprints: Stabilize before returning to balanced
5. Review strategy every 50 cycles: Adjust based on outcome success rates

---

Troubleshooting

Strategy not applied

Check if FORCE_INNOVATION or circuit breaker is overriding your setting:

grep -r "FORCE_INNOVATION" memory/evolution/

Too many repairs

Switch to harden or repair-only:

EVOLVE_STRATEGY=harden node index.js --loop

Stuck in repair loop

Circuit breaker should auto-trigger. If not, check repairLoopThreshold:

node -e "console.log(require('./src/gep/strategy').resolveStrategy())"

---

Related

• Environment Variables
• Debugging
• Common Issues