Choosing a groundedness metric when no leaderboard covers your corpus, every candidate fails your gates, and the bias you ship has to be documented instead of hidden.
I was two days from rewriting a working system to fix a deficit that did not exist. Cosine similarity was scoring vocabulary reuse, not truth.
A grid search across 15 RAG configs revealed that chunk size matters more than embedding model, overlap is not optional, and bigger parameters don't mean better recall.
I fine-tuned all-MiniLM-L6-v2 on dating profiles, flipped Spearman from -0.22 to +0.85, and found LoRA hit 96.2% of that with 0.32% of parameters.
Grid search across 16 RAG configurations reveals embedding model selection drives 26% more retrieval quality than chunk tuning.