proffesor-for-testing__agentic-qe/assets/governance/shards/learning-optimization.shard.md

Learning Optimization Domain Shard

Domain: learning-optimization Version: 1.0.0 Last Updated: 2026-02-03 Parent Constitution: .claude/guidance/constitution.md

---

Domain Rules

1. Pattern Validation Before Sharing: Learned patterns MUST be validated (minimum success rate 0.7) before cross-domain sharing; unvalidated patterns pollute other domains.

2. Experience Source Verification: Experiences used for learning MUST have verified outcomes; learning from incomplete or assumed results is prohibited.

3. Knowledge Synthesis Traceability: Synthesized knowledge MUST maintain provenance links to source experiences; orphan knowledge is not actionable.

4. Strategy A/B Testing: New strategies MUST be A/B tested against baselines before full adoption; untested strategy changes are high-risk.

5. Model Export Integrity: Exported learning models MUST include version, training data summary, and validation metrics; metadata-free exports are rejected.

6. Cross-Domain Conflict Resolution: When patterns from different domains conflict, the domain with higher confidence and more recent validation wins.

---

Quality Thresholds

Metric	Minimum	Target	Critical
Pattern Success Rate	0.7	0.85	< 0.5
Experience Verification Rate	0.9	0.98	< 0.8
Knowledge Provenance	1.0	1.0	< 0.9
Strategy A/B Confidence	0.95	0.99	< 0.9
Cross-Domain Sharing Rate	0.3	0.5	N/A
Learning Velocity	> 0	> 0.1/week	< 0

---

Invariants

INVARIANT pattern_validation_before_sharing:
  FOR ALL shared_pattern IN cross_domain_patterns:
    shared_pattern.success_rate >= 0.7 AND
    shared_pattern.validation_count >= 3 AND
    shared_pattern.last_validated IS NOT NULL

INVARIANT experience_verification:
  FOR ALL experience IN learning_experiences:
    experience.outcome_verified = true AND
    experience.verification_method IN ['test_execution', 'human_review', 'automated_check']

INVARIANT knowledge_provenance:
  FOR ALL knowledge IN synthesized_knowledge:
    knowledge.source_experiences.length >= 1 AND
    FOR ALL source IN knowledge.source_experiences:
      source.experience_id IS NOT NULL

INVARIANT strategy_ab_testing:
  FOR ALL new_strategy IN adopted_strategies:
    EXISTS ab_test WHERE
      ab_test.strategy_id = new_strategy.id AND
      ab_test.completed = true AND
      ab_test.confidence >= 0.95 AND
      ab_test.new_strategy_wins = true

INVARIANT model_export_integrity:
  FOR ALL export IN model_exports:
    export.version IS NOT NULL AND
    export.training_data_summary IS NOT NULL AND
    export.validation_metrics IS NOT NULL AND
    export.created_at IS NOT NULL

---

Patterns

Domain Source: v3/src/domains/learning-optimization/

Pattern	Location	Description
Learning Coordinator Service	services/learning-coordinator.ts	Cross-domain orchestration
Transfer Specialist Service	services/transfer-specialist.ts	Pattern transfer between domains
Metrics Optimizer Service	services/metrics-optimizer.ts	Strategy optimization
Production Intel Service	services/production-intel.ts	Production feedback integration
Learning Optimization Coordinator	coordinator.ts	Workflow orchestration

Key Interfaces: interfaces/index.ts defines LearnedPattern, Experience, Knowledge, Strategy, and related types.

---

Agent Constraints

Role	Agent ID	Permissions
Primary	qe-learning-coordinator	Full learning orchestration
Secondary	qe-pattern-learner	Pattern extraction, validation
Secondary	qe-transfer-specialist	Cross-domain pattern transfer
Secondary	qe-metrics-optimizer	Strategy optimization
Support	All domain agents	Report experiences
Readonly	qe-quality-gate	Query learning status

Forbidden Actions: No agent may share unvalidated patterns across domains.

---

Escalation Triggers

Trigger	Severity	Action
Learning velocity < 0 (regression)	CRITICAL	Escalate to Queen Coordinator, investigate
Pattern success rate < 0.5	HIGH	Quarantine pattern, escalate
Unverified experience used	HIGH	Rollback learning, escalate
Knowledge provenance broken	HIGH	Quarantine knowledge, reconstruct
A/B test confidence < 0.9	MEDIUM	Extend test duration
Cross-domain conflict unresolved	MEDIUM	Escalate to Queen Coordinator
Model export without metadata	MEDIUM	Block export, request metadata
Experience verification rate < 0.8	MEDIUM	Audit verification pipeline

---

Memory Namespace

• Namespace: qe-patterns/learning-optimization
• Retention: 180 days (patterns), 90 days (experiences)
• Contradiction Check: Enabled (critical for learning consistency)

---

Integration Points

Domain	Integration Type	Purpose
All domains	Input	Receive experiences and outcomes
All domains	Output	Share validated patterns
quality-assessment	Output	Report learning metrics
defect-intelligence	Bidirectional	Share defect patterns
coverage-analysis	Bidirectional	Share coverage patterns

---

Pattern Types

Type	Description	Sharing Scope
test-pattern	Effective test strategies	test-generation, test-execution
coverage-pattern	Gap detection strategies	coverage-analysis
defect-pattern	Defect prediction indicators	defect-intelligence
security-pattern	Security testing strategies	security-compliance
performance-pattern	Load test configurations	chaos-resilience
meta-pattern	Cross-domain optimizations	All domains

---

Experience Recording Schema

interface Experience {
  id: string;
  domain: string;
  action: string;
  state: StateSnapshot;
  result: ExperienceResult;
  outcome: {
    verified: boolean;
    verificationMethod: 'test_execution' | 'human_review' | 'automated_check';
    success: boolean;
    metrics: Record<string, number>;
  };
  timestamp: Date;
  agentId: string;
  sessionId: string;
  resourceUsage: ResourceUsage;
}

---

Strategy A/B Testing Protocol

interface ABTestConfig {
  strategyId: string;
  baselineStrategyId: string;
  trafficSplit: number; // 0.5 = 50/50
  minSampleSize: number;
  targetMetric: string;
  confidenceLevel: number; // 0.95 default
  maxDuration: string; // "7d" default
  earlyStoppingEnabled: boolean;
}

---

Cross-Domain Conflict Resolution

Priority	Factor	Weight
1	Confidence score	0.4
2	Recency of validation	0.3
3	Sample size	0.2
4	Domain authority	0.1

When patterns conflict:

1. Calculate weighted score for each pattern
2. Pattern with highest score wins
3. Losing pattern marked as superseded
4. Supersession recorded in memory with explanation

---

Learning Cycle Report Schema

interface LearningCycleReport {
  cycleId: string;
  period: {
    start: Date;
    end: Date;
  };
  experiencesRecorded: number;
  experiencesVerified: number;
  patternsLearned: number;
  patternsShared: number;
  strategiesOptimized: number;
  abTestsCompleted: number;
  learningVelocity: number;
  domainBreakdown: Record<string, {
    experiences: number;
    patterns: number;
    improvements: number;
  }>;
}

---

This shard is enforced by @claude-flow/guidance governance system.