AI Performance
Orchestration Model • Version v1.7-demo
Evaluation
NewDecision Agreement with Expert
93.7%
1,203 of 1,284 cases
Expert Override Rate
6.3%
81 overrides — the exact complement of agreement
Unsupported Recommendation Rate
1.8%
23 cases lacked a full evidence chain
Evidence Coverage
98.2%
1,261 cases fully evidence-backed
Automated Routing Accuracy
94.2%
1,210 cases routed to the correct specialist
Low Confidence Escalation
100%
all 387 below-threshold cases escalated
Current Model
NewLLM-A
- Decision Agreement: 93.7% (in production)
- Cost/case: $0.14
- Latency: 2.3 sec
Candidate Model
NewLLM-B
- Decision Agreement: 93.0% (offline evaluation)
- Cost/case: $0.08
- Latency: 1.4 sec