Unverified50% confidenceBenchmarkExact time
After the fix, the invariant that streamed output matches whole-string output holds across roughly 320,000 fuzzing chunks.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
UnverifiedRetesting with the same 20 files, 18 compiled directly; after one more iteration letting the sessions receive compiler error messages, it reached 19.67% 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.67% similarUnverifiedIn the first test case, with one partial return (queue_s=35, gen_s=9, emitted_tokens=220, shipped=false) and a policy of deadline_s=600, ceiling_tokens=4000, next_queue_s=35, next_gen_s=12, next_est_tokens=800, the result should be retry_free.64% similarUnverifiedAn evaluation suite with 1,800 test cases has long maintained a stable aggregate pass rate of 97%, but the 64-case multilingual slice within it has a pass rate of only 61%62% similarUnverifiedIn 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.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1062939API
curl https://kongchang.com/api/v1/knowledge/claims/1062939MCP
get_claim(id=1062939)