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 / C4-H2
What this task measures
The caller is on the line with Straus Dermatology for one locked request: Botox on an existing cosmetic consult. They keep repeating that ask through a waitlist offer, a callback, or a different errand. We included it as a hard cosmetic concierge case because quoted cosmetic policy, deposits, and the cooling-off window have to be spoken before a booking is taken. The agent succeeds only if it stays on that request, reaches the outcome by following the identity then cosmetic concierge path, and leaves the record untouched: no appointment, ticket, or note the caller never asked for.
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 $0.301
0:00 / 3:23 · audio · trace 801906 · 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
+ from: reception
- next_intent: cosmetic
+ to: identity
Output
None recorded.
None recorded.
Parameters
- dob: 1972-06-30
- first_name: Maria
- last_name: Alvarez
Output
None recorded.
None recorded.
Parameters
dob: 1972-06-30
dob: 1972-06-30
full_name: Maria Alvarez
full_name: Maria Alvarez
+ second_factor: 10016
Output
- data: {"verified":true}ok: true
ok: true
Parameters
None recorded.
None recorded.
Output
+ ok: true
Parameters
+ from: identity
+ to: cosmetic
Output
None recorded.
None recorded.
Parameters
service: botox
service: botox
Output
+ ok: true
Parameters
+ amount_cents: 30000
Output
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. Missing: identify_patient. 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_identity → transfer_to_cosmetic
Actual path
transfer_to_identity → transfer_to_cosmetic
The actual hangup matches the expected hangup.
Locked caller script. This is the technical definition. Copy it as written.
You are an existing Straus patient. You already have a cosmetic consult booked at Park Avenue, and you want to keep that exact visit. Answer identity questions once, then get back to why you called. Ask three things, in this order: what Botox would cost; whether you can talk about Botox at the consult you already have; and after you hear the price, say that is more than you can pay all at once and ask whether there is a way to spread the cost out. Stay on the line until you have a price, confirmation that Botox can be discussed at that existing visit, and an answer about spreading the cost. Then thank them and end the call. Do not ask to book a new consult. Do not ask to reschedule or cancel the visit you already have. Do not ask for a street address, floor, suite, or location id. Do not ask for a callback, to speak to a person, about a medical bill, the patient portal, a waitlist, or a text. Answer identity first if they ask who you are. When you ask to spread the cost, ask whether CareCredit is available and stay until they have answered that from their system. Do not ask them to text you an appointment confirmation.