How rigorous is your team's LLM evaluation right now?

Self-assessed — we're trying to map where most teams actually are, not where they want to be.

• We don't have one. Vibes-based.
• Ad-hoc spot checks when something breaks
• A labeled eval set we run manually
• Automated evals in CI on every change
• Layered: deterministic + LLM-judge + regression set