Your Multi-Model Fallback Fails Together

Why a second LLM provider is not a second opinion, and how to measure whether your redundancy is real.