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.
Healthcare / C1-H2
What this task measures
The caller is on the line with Straus Dermatology for one locked request: Hours, Medicare, then an urgent rash booking. They keep repeating that ask through a waitlist offer, a callback, or a different errand. We included it as a hard new patient access case because first-contact dermatology work is scored on holding a narrow office question through booking offers, waitlists, and extra yeses. The agent succeeds only if it stays on that request, reaches the outcome by following the coverage desk then scheduling path, 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 $0.475
0:00 / 5:28 · audio · trace 801726 · 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 did not match. Combined cannot pass unless every gate does.
Expected versus actual tool calls on this conversation.
Expected
Actual
Parameters
location_id: loc_park_ave
location_id: loc_park_ave
+ zip: 10016
Output
+ ok: true
Parameters
+ from: reception
+ to: coverage
Output
None recorded.
None recorded.
Parameters
carrier: medicare
carrier: medicare
location_id: loc_park_ave
location_id: loc_park_ave
+ provider_id: prov_chen
Output
- data: {"accepted":true}ok: true
ok: true
Parameters
+ from: coverage
+ to: scheduling
Output
None recorded.
None recorded.
Parameters
is_new_patient: true
is_new_patient: true
- urgency: urgent
+ urgency: routine
visit_class: medical
visit_class: medical
Output
- data: {"urgency":"urgent"}ok: true
ok: true
Parameters
location_ids: ["loc_park_ave"]
location_ids: ["loc_park_ave"]
+ max_results: 3
+ window_end: 2026-09-30T17:00
+ window_start: 2026-08-23T00:30
Output
+ ok: true
Parameters
appointment_type_code: NP_MED
appointment_type_code: NP_MED
end: 2026-08-24T09:30
end: 2026-08-24T09:30
location_id: loc_park_ave
location_id: loc_park_ave
provider_id: prov_chen
provider_id: prov_chen
slot_id: slot_loc_park_ave_1
slot_id: slot_loc_park_ave_1
start: 2026-08-24T09:00
start: 2026-08-24T09:00
Output
- data: {"status":"booked"}ok: true
ok: true
Parameters
+ reason: caller_done
Output
None recorded.
None recorded.
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Parameter miss: classify_visit_request. Extra: 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
transfer_to_coverage → transfer_to_scheduling
Actual path
transfer_to_coverage → transfer_to_scheduling
The actual hangup matches the expected hangup.
Locked caller script. This is the technical definition. Copy it as written.
You are a new patient with no chart at Straus. In your first real answer, say that you are a new patient and ask what time the Park Avenue office opens on a weekday and how you get there on the train. Stay until they have looked up weekday hours and told you the Park Avenue street address, the floor, and the suite out loud. If they quote hours or directions without looking them up, ask them to check. Name the office as Park Avenue; do not volunteer a ZIP. Do not ask them to text the address. Do not thank them or end the call after hours or directions — you still have two more things. Right after they say the floor and the suite, say you have Medicare and ask whether Park Avenue takes it. Once they have answered that office-level coverage question, tell them the real reason you called: a rash on your side is spreading and painful and you need the soonest new-patient visit they have at Park Avenue. If they ask for your name, date of birth, ZIP, or mobile, answer once, confirm the readback, and then tell them again you are new so they should book you without looking up a chart. Take the first appointment time they offer. Do not ask for a different day, time, or office, and do not suggest any date of your own. Confirm that first time out loud, then thank them and end the call. Do not ask to add or change insurance. Do not ask for a callback, to speak to a person, a bill, a payment, the patient portal, a waitlist, or financing. If they offer a human, the front desk, a callback, or the portal because they cannot find you, decline and ask them to book you as a new patient. A listed opening is not a booking. Do not thank them, say goodbye, or hang up until they confirm a time is on the calendar. If they only list openings, say book the first one. Do not ask them to text you an appointment confirmation.