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 / C2-H2
What this task measures
The caller is on the line with Straus Dermatology for one locked request: Allergy shot for a child. They keep repeating that ask through a waitlist offer, a callback, or a different errand. We included it as a hard appointment management case because calendar writes are a core front-desk job, and they fail in characteristic ways: the wrong visit moves, identity is skipped, or a cancel becomes a reschedule. The agent succeeds only if it stays on that request, reaches the outcome by following the identity then 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 $0.727
0:00 / 8:08 · audio · trace 801786 · 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 handoff path 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
- next_intent: coverage
Output
None recorded.
None recorded.
Parameters
- dob: 2016-03-22
- first_name: Leo
- last_name: Park
Output
None recorded.
None recorded.
Parameters
- dob: 2016-03-22
- full_name: Leo Park
Output
- data: {"verified":true}- ok: true
Parameters
None recorded.
None recorded.
Output
- ok: true
Parameters
+ from: reception
+ to: coverage
Output
None recorded.
None recorded.
Parameters
carrier: aetna
carrier: aetna
location_id: loc_brooklyn_heights
location_id: loc_brooklyn_heights
+ provider_id: prov_ruiz
Output
- data: {"accepted":true}ok: true
ok: true
Parameters
+ from: coverage
+ to: scheduling
Output
None recorded.
None recorded.
Parameters
location_id: loc_brooklyn_heights
location_id: loc_brooklyn_heights
+ zip: 11201
Output
+ ok: true
Parameters
location_id: loc_brooklyn_heights
location_id: loc_brooklyn_heights
service: allergy_shot
service: allergy_shot
+ window_end: 2026-08-23T17:00
+ window_start: 2026-08-23T09:00
Output
+ error_code: NO_AVAILABILITY
+ ok: false
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. Missing: transfer_to_identity, identify_patient, verify_identity, get_patient_summary. 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 through the shared prefix, then red from the first wrong hop.
Expected path
transfer_to_identity → transfer_to_coverage → transfer_to_scheduling
Actual path
transfer_to_coverage → transfer_to_scheduling
Rows that differ between the expected hangup and the actual hangup.
missing · appointments · 4
Expected
Actual
- appointment_type_code: ALLERGY_ALLERGY_SHOT
- description: Allergy service: allergy_shot
- end: 2026-08-24T10:00
- id: 4
- location_id: loc_brooklyn_heights
- patient_id: pat_leo_park
- provider_id: prov_ruiz
- start: 2026-08-24T09:00
- status: booked
Locked caller script. This is the technical definition. Copy it as written.
You are a parent calling on behalf of your son. He is an existing patient and he is due for his next allergy shot at Brooklyn Heights. When they ask who the patient is, give his name and date of birth once, say you are his father if they ask who is calling, then continue. Do not repeat his name or date of birth unless they say it did not match. Ask whether Brooklyn Heights takes his Aetna before you schedule anything. Ask for the Brooklyn Heights street address, floor, and suite, and ask how long you have to stay after the shot so you can plan pickup from school. If they ask what day or search window you want, or yes or no to search a date range, say first available, not yes. If they offer a callback, decline. Do not suggest a date, date range, or time of your own, and do not ask for a different office. Take the first daytime appointment they offer at Brooklyn Heights. If they offer a time outside office hours, such as midnight, say that does not sound right and take the first daytime time they offer next. Confirm that time out loud, then thank them and end the call. Do not move, reschedule, or cancel any other appointment. Do not ask for a regular medical visit. Do not ask to add insurance, for a callback, to speak to a person, a bill, a payment, the patient portal, a waitlist, a text, or financing. 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. If they start to book a regular medical visit, stop them and say this is his allergy shot, not a regular appointment. Stay until the shot is scheduled. Do not ask them to text you an appointment confirmation.