Unverified50% confidenceBenchmarkExact time
Tested with an 8B local model running on CPU, it correctly selected the tool 12 times out of 12 questions.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn testing, all nine inputs of varying difficulty were routed through demo-smart, and the actually selected models matched expectations in every case.72% similarUnverifiedThe test was conducted on October 12 and 13, using a total of five design tasks and completely identical prompt inputs across seven tools.70% similarUnverifiedThe test was built on the n8n workflow automation platform.70% similarUnverifiedThe n8n workflow provided in the article contains five synthetic test leads, forming a fixed test fixture that demonstrates the full process from loading, validation, and scoring to routing and assertion.69% similarUnverifiedAn 8-input controlled experiment showed that human-guided recursive refinement can make every input satisfy all 7 requirements, whereas under the parallel Oracle@8 scheme, on average only 3 can be satisfied.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1054873API
curl https://kongchang.com/api/v1/knowledge/claims/1054873MCP
get_claim(id=1054873)