Unverified50% confidenceFactExact time
In streaming LLM APIs, some relay stations silently insert data: [DONE] to end early, or stuff error messages such as {"error": "rate limit exceeded"} into JSON disguised as model answers.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified切换到推理模型时因服务端严格校验tool_calls参数必须是合法JSON而返回400错误，一条格式错误的消息会毒化整个对话历史导致后续请求全部失败70% similarUnverifiedOnce final JSON parsing fails, subsequent semantic checks are skipped entirely, which is the technical root cause of the 'task completed yet zero score' phenomenon.67% similarUnverifiedMost of what goes wrong in LLM features is not in the model itself but in the surrounding code, such as building the messages list, selecting the model ID, parsing JSON, handling an empty choices list, retrying on 429s, and counting tokens.66% similarUnverified工具调用的可靠性高度依赖模型输出严格遵循预定义的 JSON Schema，任何字段类型偏差、缺失或多余都可能导致框架层的解析失败66% similarUnverifiedThe most common model failure is returning something that is 'almost JSON,' such as extra commas, markdown code fences, verbose preambles, or writing the number 3 as the word 'three,' with the AI node reporting success while the workflow crashes only several steps later when parsing fails.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1030994API
curl https://kongchang.com/api/v1/knowledge/claims/1030994MCP
get_claim(id=1030994)