Unverified50% confidenceBenchmarkExact time
In the simple scenario of the benchmark (1,024 memories), the three rules of deleting the source, deleting similar content, and invalidating one hop left 87% to 94% contamination residue.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedIn the ablation experiment, replacing the gate with the naïve check commonly used in unconstrained systems (fewer visible errors in recent trajectories), the same loop admitted 309 harmful changes and left the miss rate above 10% in 49 out of 90 runs.72% similarUnverified19.4% (6,251 articles) of duplicate document contamination exists in the naive test set split.69% similarUnverifiedIn memory tests, retaining the full context allowed the model to accurately locate specific details mentioned by students 95% of the time, whereas after summary compression the accuracy dropped sharply to 32%.68% similarUnverifiedThe benchmark started with 15,606 raw scoring records, which after cleaning yielded 351 qualifying cases, with placeholder scores accounting for 31% of all scoring rows being excluded.68% similarUnverifiedAfter disabling schema constraints and authentication, the erroneous backend dispatch rate spiked from 0% to 95.28% in replay/test evaluations.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1051851API
curl https://kongchang.com/api/v1/knowledge/claims/1051851MCP
get_claim(id=1051851)