Method
Golden set / benchmark
Curated reference questions or labels for evaluation and regression.
42 cases from 29 companies apply Golden set / benchmark. 2 were checked in an online A/B test; 22 report results in production.
Typical tasks: Evaluate models, Answer questions over data, Rank search results, Label data with LLMs and 12 more. Companies: DoorDash, Ramp, Databricks, Netflix, Grab, Yandex and 23 more. Often combined with: LLM-as-a-judge, Human in the loop, Semantic layer, Agent evaluation, Text-to-SQL and 81 more.







