Unverified50% confidenceFactExact time
After introducing the dispute mechanism, the same model's success rate became 20/20; the control scenario (an app in the foreground but no one touching it) still maintained 20/20 with zero model calls.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedResearch shows that different prompt strategies on the same model can lead to differences in recall of more than 20 percentage points.72% similarUnverifiedOn a real Android device, the same 4B small model jumped from a 0/20 to a 20/20 success rate simply by changing the policy of "when to consult the model".71% similarUnverified在SWE-bench Full版本上，多数模型的通过率仍低于20%71% similarUnverifiedA small reviewer model costs about 1/20 per call of a mid-tier model (at current list prices).70% similarUnverifiedIn the scoring setup, only the lexical component, which accounts for 20% of the weight, is visible during the development phase, forming a partially observable objective.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1092971API
curl https://kongchang.com/api/v1/knowledge/claims/1092971MCP
get_claim(id=1092971)