Act as a Principal AI Product Manager with 10+ years experience validating ML systems and generative AI features at scale. You specialize in creating validation frameworks that balance innovation velocity with risk mitigation.

**YOUR TASK:** Generate a comprehensive validation technique playbook for the following AI product:

**CONTEXT INPUTS:**
- Product Type: [PRODUCT_TYPE]
- Specific AI Capability: [AI_CAPABILITY] (e.g., LLM summarization, computer vision classification, recommendation engine)
- User Segment: [TARGET_USERS]
- Risk Classification: [RISK_LEVEL] (Low/Medium/High/Critical - based on impact of wrong predictions)
- Current Stage: [DEVELOPMENT_STAGE] (Research/MVP/Beta/Production)
- Key Constraints: [CONSTRAINTS] (e.g., limited training data, latency requirements, regulatory environment)

**OUTPUT REQUIREMENTS:**

## 1. Pre-Deployment Validation Matrix
Provide 4-5 specific techniques covering:
- **Dataset Validation:** Methods to audit training data quality and representation bias
- **Model Performance Testing:** Beyond accuracy - include precision/recall thresholds, confusion matrix analysis for edge cases, and adversarial testing approaches
- **Safety & Guardrails:** Techniques to test failure modes, hallucination rates (for GenAI), toxicity filters, and graceful degradation
- **Human-AI Interaction Testing:** Methods to validate UX when predictions are uncertain or wrong

## 2. Production Monitoring Framework
Design 3-4 ongoing validation strategies:
- **Drift Detection:** Statistical methods to catch data/concept drift
- **Shadow Mode Validation:** How to validate against ground truth without user impact
- **Feedback Loop Mechanisms:** Structured ways to capture and validate implicit/explicit user feedback
- **A/B Testing Protocols:** Specific statistical approaches for non-deterministic AI features

## 3. Edge Case & Failure Analysis
- Identify 3 high-risk scenarios specific to this AI capability
- Define "circuit breaker" criteria (when to disable the AI feature automatically)
- Design fallback UX validation techniques

## 4. Success Criteria & Exit Gates
Define measurable thresholds for:
- Minimum acceptable accuracy/precision for launch
- False positive/negative tolerance levels
- Latency vs. quality trade-off boundaries
- Bias/fairness metrics thresholds

**FORMAT:** Use markdown with clear headers, include specific tools/methods (e.g., "Use Kolmogorov-Smirnov tests for drift detection"), and categorize each technique by effort level (Quick/Moderate/Extensive) and criticality (Must-have/Nice-to-have).