Launched August 18, 2026, each Tuned Evaluator is a versioned judge for one objective; LangChain maintains the prompt and post-trained model. Perceived Error flags conversations where users likely experienced errors or unresolved outcomes, billed at 0.01 LCU per successful run on Plus and Cloud Enterprise US plans.