Unverified50% confidenceFactExact time
The error criterion accepted 17 runs (13 of which were pool selections) with failure probabilities as low as 0.38 to 0.89 times the reference value; another 181 runs exhausted the budget.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe defect probability calibration method reduced the Brier score in 13 of 15 model-application configurations.76% similarUnverifiedIn a benchmark of four enterprise agent architectures under 100 injected fault conditions (transient network timeouts, schema mismatches, empty search results), the naive self-reflection loop had an unconstrained runaway rate of 42.0%, median token consumption of 148,500 for failed tasks, mean time to failure of 18.4 minutes, and per-incident financial risk of $12.40.75% similarUnverifiedIn the experiments, the task resolution rate was 17% for zero-shot generation, 27% for generic self-correction, 23% for error-only feedback, and 90% for diagnostic counterexample feedback (A-CEGIS).74% similarUnverifiedUnder faults, the 120B model did not retain a higher proportion of its fault-free score than the 20B model; the p = 0.96 for the fault-tolerance-rate model indicates that the scale difference is not statistically significant.73% similarUnverifiedThe criteria are divided into four families—foundational, premature disclosure, goal deviation, and information omission—with 19 failure modes in total, of which 18 were triggered in actual testing.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1130859API
curl https://kongchang.com/api/v1/knowledge/claims/1130859MCP
get_claim(id=1130859)