Unverified50% confidenceFactExact time
In the third round of routing tests, using the weak model DeepSeek V4.1-Flash, across 480 routing judgments, all failures in the pipeline group manifested as the model refusing to trigger rather than choosing the wrong skill; after removing the boundary negation clause from the description, such failures dropped from 10 to 0.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
UnverifiedIn the fourth round of experiments, trap questions narrowed the gap from 6.6 percentage points in the third round to 2.7 percentage points, and the failure mode became the model choosing the wrong neighboring skill rather than refusing to trigger.72% similarUnverifiedThe study employs a triple-judgment mechanism: fail_fast, endstate, and diligence, with each multi-turn test broken down into three mutually exclusive dimensions.68% similarUnverifiedIn benchmark testing, execution failures and model decisions should be counted separately to avoid misinterpreting a single quota overrun as a deficiency in model capability.68% similarUnverifiedIn a parse-then-extract two-stage pipeline, errors accumulate—table reconstruction errors in the parsing stage propagate to extraction results, causing end-to-end accuracy to fall below the independent performance of each individual module.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1027065API
curl https://kongchang.com/api/v1/knowledge/claims/1027065MCP
get_claim(id=1027065)