[KongchangAI]
Unverified50% confidenceBenchmarkExact time

The study shows that user satisfaction carries almost no information about whether a task was truly completed, and conversations judged as 'satisfactory' are decorrelated from actual task success.

1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen

Sources

GAUGE Study: When You Can't Trust LLM-as-a-Judge for Agent Evaluation

Related Entities

GAUGEconcept

Related Claims

UnverifiedUser satisfaction equals actual experience minus expectations.68% similarUnverified作业分数衡量任务完成质量，而考试分数衡量知识内化程度，二者存在关键差异63% similarUnverified产品的成败仍然取决于想法本身的质量、对用户痛点的洞察以及后续的运营和迭代能力63% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/919800
API
curl https://kongchang.com/api/v1/knowledge/claims/919800
MCP
get_claim(id=919800)