Unverified50% confidenceFactExact time
The research team conducted a retrospective ablation analysis across 12 models with a total of 121,680 valid trials, of which 79,853 failed the executability check.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/9/2026
First Seen
Valid until: 1/7/2027
Sources
Related Entities
Related Claims
VerifiedA 2016 survey by Nature showed that more than 70% of researchers had attempted to replicate others' experiments and failed.70% similarUnverifiedThe study conducted paired comparison experiments of 10,000 runs each, validating two separation phenomena, exact fixed-point recovery, non-monotonic stability regions, and task dependence.69% similarUnverifiedThe researchers confirmed 91 failures through manual verification and categorized them into 7 failure loci.67% similarUnverified该测试样本规模相对有限（5个模型、13个代码库），结论有待更大规模验证66% similarUnverifiedThe study conducted more than ten million trials, covering twelve LLMs from four model families across eight domains.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/991242API
curl https://kongchang.com/api/v1/knowledge/claims/991242MCP
get_claim(id=991242)