Resolution integrity
Match the agent’s claim of completion to the tool state and required workflow steps.
Customer support
Evaluate support agents against the queue rules, account workflows, escalation paths, and caller conditions that determine real resolution.
Why this needs evaluation
What Voxeval proves
Match the agent’s claim of completion to the tool state and required workflow steps.
Verify destination, availability, context preservation, and the moment escalation should occur.
Re-run key flows with interruptions, repetition, frustration, silence, and noisy audio.
Evaluation sequence
Model the intent and acceptable outcomes.
Test resolution and escalation branches.
Gate releases on the failures customers would feel.