How to evaluate LLM

For anyone selecting an AI development partner, I would look beyond the model names in the proposal. The more useful questions concern data access, integration ownership, evaluation methodology, security, observability, deployment, intellectual property and support after launch. A team should be able to explain how the system behaves when retrieval fails, a model changes or an external API becomes unavailable. We apply the same criteria when scoping projects at AI Development Services. What would you add to this checklist?