Unverified50% confidenceEventExact time
An end-to-end test asserted that the model's answer must contain the literal string "Ledger freeze ends Friday", and it failed in four different ways across four runs (lowercase, Markdown bold, an invisible narrow no-break space, and paraphrased to "ends on Friday"), even though the data was correct every time.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFor ambiguous time-expression scenarios (such as 'next Friday' or 'tomorrow evening'), tests must be run multiple times, because a single clean result tells you almost nothing.65% similarUnverified终止条件设计不当会导致无限循环（反复调用工具无法收敛）或过早终止（信息不足时就回复）两类故障64% similarUnverifiedDuring testing, the author fixed the problem of "long answers being cut off mid-way" by increasing the pause detection time and the maximum answer length.63% similarUnverifiedIn the control experiment, the same 4B model exhausted all six rounds and failed under the old framework, while after the fix it completed in five rounds (pytest→read→fix→re-verify, correct).63% similarUnverifiedIn official tests, after removing instructions such as 'think carefully before answering,' Opus 5.5 began responding faster with no noticeable drop in quality.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1062579API
curl https://kongchang.com/api/v1/knowledge/claims/1062579MCP
get_claim(id=1062579)