Unverified50% confidenceFactExact time
On the weekly time window, confirmatory tests detected no ranking information beyond pure code features, i.e., after controlling for price features, LLM text judgments provided no additional predictive power.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Claims
UnverifiedPre-specified diagnostics showed that date information played no measurable role in classification, while models varied greatly in their sensitivity to source order.68% similarUnverifiedVector retrieval performs terribly at looking up exact identifiers (such as contract numbers, customer names, date ranges, and material codes), because embedding models compress text into dense vectors, and similarity ranking has no binding force for exact equality.67% similarUnverifiedLLM-based code auditing is limited by the context window, may produce hallucinated false positives, and struggles to guarantee reproducible results.67% similarUnverifiedEvaluation in financial scenarios cannot look only at text fluency; it must test the model's performance on number extraction, computational consistency, and the completeness of the logical chain.65% similarUnverifiedVector retrieval excels at fuzzy semantic matching but is insensitive to precise conditions (such as time ranges, user IDs, and status enum values), and its recall results lack explainable filtering criteria.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1154080API
curl https://kongchang.com/api/v1/knowledge/claims/1154080MCP
get_claim(id=1154080)