What happens after the patient asks for help?
Patient message → safety assessment → staff handoff → documented resolution. ClinEval records each stage separately so an AI response cannot stand in for completed follow-up.
- Missed urgent cases
- Urgent reference cases classified as routine. Missed required escalations are reported separately.
- Unnecessary escalations
- Escalations among cases where the reference judgment does not require staff review.
- Handoff completion
- Required handoffs with a recorded staff review. Assignment alone does not count as acceptance.
- Resolution time
- Time from the patient message to a recorded follow-up outcome. Reports include unresolved cases and the measured sample size.
Current evidence: local software checks
Clinician reference review and a full run with a live model remain pending. Synthetic workflow checks do not establish clinical performance or production readiness.
Staff can record care received, completed follow-up, refusal, or an unreachable patient. These records document staff attestations. They do not independently verify medical outcomes.
Discuss clinician reviewSix Dimensions of Clinical Quality
Each dimension is weighted by clinical importance. Safety detection carries the highest weight because missing an emergency has irreversible consequences.
Safety Detection
30% weightEvaluates emergency and urgent situation recognition with asymmetric error weighting—missing emergencies is penalized heavily.
Triage Accuracy
25% weightMeasures accuracy of clinical classification, severity assessment, and module triggering across conditions.
Escalation Quality
20% weightAssesses human handoff decisions—timing, accuracy, and the critical balance between false positives and missed escalations.
Response Appropriateness
15% weightEvaluates clinical accuracy, guideline adherence, tone appropriateness, and absence of harmful content.
Confidence Calibration
5% weightMeasures reliability of confidence scores—a well-calibrated system should be right 70% of the time when it reports 70% confidence.
Contextual Coherence
5% weightTests multi-turn consistency, RAG context utilization, and proper use of patient history.
Versioned Test Cases
112 versioned response cases cover emergency detection, triage, adversarial inputs, and clinical scenarios. The separate workflow suite uses five synthetic scenarios. Clinician reference review remains pending.
Emergency Detection
Explicit, implicit, and multilingual emergency cases
Triage Scenarios
Clinical classification, severity, and module selection
Escalation Decisions
Required handoffs, non-response, and escalation timing
Response Quality
Required content, tone, and prohibited guidance
Adversarial Inputs
Prompt injection, misleading symptoms, and unsafe requests
Evaluation Methodology
ClinEval applies asymmetric error weights because missed emergencies have greater clinical risk.
Asymmetric Weighting
The response benchmark applies a 10x penalty to an emergency classified as routine. This weighting is a scoring policy, not clinical validation.
Latency Requirements
The response benchmark uses a safety-detection target below 100 milliseconds. This target does not measure staff response or resolution time.
Baseline Tracking
Teams can compare completed reports against an approved baseline. Automated release gating is not active.
Evidence Across the Care Workflow
The local workflow suite exercises conversation handling, safety assessment, escalation records, staff review, and follow-up records. The language-model response and notification delivery use test substitutes.
Medical Safety Engine
Emergency and urgent detection with multilingual support (English, Hindi, code-switching).
Triage System
Module-based classification with confidence scoring and escalation recommendations.
Escalation Rules
Context-aware human handoff decisions with SLA tracking and notification routing.
