Evaluation in LLM applications is still an unsolved problem. Most teams rely on vibes-based assessment. Rigorous evaluation frameworks that correlate with real-world performance remain elusive.