Expired50% confidenceFactExact time
Trajectory Eval is used to analyze differences in Agent behavior across different paths; for example, when an Agent takes path B and the signal drops, the root cause may be an incorrect component call sequence.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
LLM Observability & Eval in Practice: A Complete Guide to Monitoring and Evaluating Large Models in Production
bilibili量子菠萝_6/8/2026
Related Claims
Unverifiedinaturalist-clumper的聚合算法借鉴了轨迹分析中的停留点检测思想，按时间排序遍历观察记录，使用Haversine公式计算地理距离64% similarUnverifiedHaving the model roll forward a number of steps and then roll backward the same number of steps should ideally return to the starting point, and the discrepancy between the actual return position and the starting point serves as a self-supervised proxy metric for unobservable rollout error.63% similarUnverifiedREINFORCE估计器的方差极高，因为一条轨迹中的每个动作都乘以整条轨迹的总回报62% similarUnverified在确定性高的表单填写场景中，脚本处理高频确定性路径、Agent作为fallback处理边缘情况的混合架构优于纯LLM驱动方案61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/49927API
curl https://kongchang.com/api/v1/knowledge/claims/49927MCP
get_claim(id=49927)