2 docs tagged with "observability"

Models fail silently. They keep returning confident, plausible-looking answers that are simply wrong, with no crash, no error, nothing that would trip a standard software alert — the only way to catch this is to watch the inputs and predictions continuously, because the ground truth needed to know an answer was wrong often arrives late, or never.

LangSmith tracing records every step a LangChain/LangGraph run takes — each model call, tool call, and retriever lookup — as a nested tree you can inspect after the fact. It's the fastest way to find which step in a multi-step chain produced a bad answer.