Overall score
99%
High Assurance threshold
Ground Truth Evaluation
Measure whether the system is useful, grounded, and safe.
Deterministic checks are primary. Model-based judging is supplemental and clearly labeled. Every result links to a trace and versioned expected outcome.
Cases
28
Core · advanced · adversarial
Deterministic pass
28 / 28
0 review cases
Regressions
0
Compared with fixture v1.0
Quality dimensions
Current deterministic replay suite
Latest run
Fixture version CTI-EVAL-1.0
99%
Approved
Grounding, safety, and utility gates passed.
Evidence threshold≥80% 94%
Grounding threshold≥90% 98%
Material contradictions0 0
Unsafe SQL accepted0 0
Private chain-of-thought stored0 0