Unverified50% confidenceSolutionExact time
In a database containing 366 ga_sessions_YYYYMMDD date-sharded tables, providing the tables one by one to the model filled all 12 slots with random dates and the model refused to answer; after collapsing each table family into a wildcard table (ga_sessions_* with _TABLE_SUFFIX filtering), most refusals disappeared.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMost 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.59% similarUnverifiedThe time saved by the hexagonal architecture in the database extension scenario was not enough to offset the cumulative time gap accumulated across the preceding fifteen-feature sequence.58% similarUnverifiedIn a RAG system, if the document parsing stage loses summary rows due to the merged-cell problem, that information will not exist in the vector database, and subsequent retrieval and generation cannot restore it.56% similarUnverified在RAG流程返回大量候选文档片段时，可先按相关性排序再截取前3-5项传入大模型以控制输入长度55% similarUnverifiedCommon RAG mistakes include improper chunking strategies, mismatch between the embedding model and queries, and lack of reranking on retrieved results.55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1092539API
curl https://kongchang.com/api/v1/knowledge/claims/1092539MCP
get_claim(id=1092539)