Working paper / 2026
Multi-Industry Voice Agent Simulation Bench
Abstract
Multi-Industry Voice Agent Simulation Bench (MIVAS) is an indicator of voice AI performance across economic sectors. MIVAS is comprised of tasks and verifiers with granular rewards executed in production-grade multi-agent RL environments across industries where voice AI is being adopted the fastest. Each task places a model harness inside a multi-agent graph with production-length prompts, tools, and isolated state. Deterministic verifiers score tool use, handoffs, and final database state; pass@1 and passk measure single-run capability and repeated-run reliability.
Clone the repository, pair a model harness with an industry task suite, and reproduce the evaluation environment.
Legal / C1-H3
What this task measures
The caller is on the line with Halverson & Reed for one locked request: Medical emergency on the line. They keep repeating that ask through a callback offer, a message left for someone else, or advice on a different matter. We included it as a hard reception and routing case because intake has to wait until the caller is classified; messages, wrong numbers, and represented parties are everyday reception work. The agent succeeds only if it stays on that request, reaches the outcome by staying on reception, and leaves the record exactly as this case requires: the right row written or cleared, and nothing extra.
One model harness at a time, with a representative pass and fail conversation when both exist. OpenAI Realtime 2.1 is the default view.
Recording
Conversation cost $0.434
0:00 / 5:56 · audio · trace 803821 · click a span to play, drag to scrub, scroll to pan
Audio
Speech Audio
Agent Audio
Handoffs
Tool Calls
Errors
Transcript
How a conversation passes
Combined pass is the logical and of three deterministic gates. Tools, handoff path, and final database state must all pass on the same conversation. One miss fails the call.
Combined
Fail
This sample fails because tools and final database state did not match. Combined cannot pass unless every gate does.
Expected versus actual tool calls on this conversation.
Expected
Actual
Parameters
full_name: Inez Calderon
full_name: Inez Calderon
+ phone: 9045550317
Output
+ ok: true
Parameters
+ from: reception
+ to: screening
Output
None recorded.
None recorded.
Parameters
opposing_party: Sutter Cold Storage
opposing_party: Sutter Cold Storage
Output
+ ok: true
Parameters
practice_area: workers_comp
practice_area: workers_comp
Output
+ ok: true
Parameters
practice_area: workers_comp
practice_area: workers_comp
state: FL
state: FL
Output
+ ok: true
Parameters
incident_date: 2026-07-28
incident_date: 2026-07-28
practice_area: workers_comp
practice_area: workers_comp
state: FL
state: FL
Output
+ ok: true
Parameters
+ from: screening
+ to: intake
Output
None recorded.
None recorded.
Parameters
incident_date: 2026-07-28
incident_date: 2026-07-28
practice_area: workers_comp
practice_area: workers_comp
state: FL
state: FL
+ summary: Caller reports collapsing at work at Sutter Cold Storage, was told they stopped breathing, and was taken for medical treatment.
Output
+ ok: true
Parameters
- channel: email
+ channel: sms
Output
+ ok: true
Parameters
+ reason: Caller received intake packet by SMS and does not want to schedule or provide further information.
Output
None recorded.
None recorded.
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Parameter miss: send_intake_packet. Extra: transfer_to_screening, transfer_to_intake, end_call.
Full industry architecture, same graph as the industry page. The left diagram marks the expected specialist path in green. The right diagram marks the path this conversation actually took in green, because it matched.
Expected path
stay on reception
Actual path
transfer_to_screening → transfer_to_intake
Rows that differ between the expected hangup and the actual hangup.
extra · documents · 1
Expected
Actual
+ caller_id: c_new
+ id: 1
+ kind: intake_packet
missing · documents · {"caller_id": "c_new", "kind": "intake_packet", "target": "email"}
Expected
Actual
- caller_id: c_new
- kind: intake_packet
intakes · 1
Expected
Actual
- summary: Intake summary.
+ summary: Caller reports collapsing at work at Sutter Cold Storage, was told they stopped breathing, and was taken for medical treatment.
Locked caller script. This is the technical definition. Copy it as written.
Open with exactly: "I collapsed on the job last Tuesday. They said I stopped breathing. I need a lawyer." This is your own workplace injury, already treated, not a live emergency. You have two jobs on this call: they check the other side, the matter type, Florida, and the filing deadline, and they record the intake and send the new-client packet. Stay until they confirm the intake is on the file and the packet is sent. A spoken 'we'll take care of it' without those writes is not enough. Then thank and end. Give your name and number only if asked. Do not volunteer the mill name, the state, or the date unless asked. When asked who this would be against, say exactly: "It's Sutter Cold Storage." When asked which state, say exactly: "In Florida." When asked when, say exactly: "It was July twenty eighth, twenty twenty six." When asked what type of matter this is, say exactly: "It's a workers' compensation claim. I was on the job." If they tell you to hang up and call 911, say exactly: "I already got care last Tuesday. I need a lawyer for the workplace injury." If offered the packet by email or text, say exactly: "Email is fine." If offered an appointment, say exactly: "Not yet, let me read the paperwork first." Do not ask for a person. Once they confirm the intake is on the file and the packet is sent, say exactly: "Thank you, that's all I needed." and end the call.
