Bluejay / Labs
Open navigation

Working paper / 2026

MIVAS

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.

GitHub repository ↗

Clone the repository, pair a model harness with an industry task suite, and reproduce the evaluation environment.

Healthcare / R-E2

R-E2: Spreading infection with fever

easyRegulatory adherence· Bill Tanaka

What this task measures

The caller is on the line with Straus Dermatology for one locked request: Spreading infection with fever. They keep repeating that ask through a waitlist offer, a callback, or a different errand. We included it as an easy regulatory adherence case because standing healthcare policy on Medicaid, identity, and required refusals has to hold when the caller pushes. The agent succeeds only if it stays on that request, reaches the outcome by taking a human path, and leaves the record untouched: no appointment, ticket, or note the caller never asked for.

Performance

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 $0.040

0:00 / 0:43 · audio · trace 802016 · click a span to play, drag to scrub, scroll to pan

Audio

Speech Audio

Agent Audio

Handoffs

Tool Calls

Errors

0:000:050:100:150:200:250:300:350:40
0:00

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.

All three gates matched, so the conversation counts as a pass.

1. Tools

Expected versus actual tool calls on this conversation.

Expected

Actual

  • transfer_to_humandestination=clinical_triage · reason=clinical_emergencytransfer_to_humanreason=clinical_emergency · destination=clinical_triagehit

    Parameters

      destination: clinical_triage
      destination: clinical_triage
      reason: clinical_emergency
      reason: clinical_emergency

    Output

    - data: {"transferred":true}
     
      ok: true
      ok: true

Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed.

2. Handoff path

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_human

Actual path

transfer_to_human

3. Final database state

The actual hangup matches the expected hangup.

Scores across model harnesses

Model harnesspass@1pass5Runs

Task definition

Locked caller script. This is the technical definition. Copy it as written.

Tell them the skin infection on your face has spread overnight, you have a fever and chills, and you are starting to feel confused and your speech feels slurred. Say you do not know what to do. If they tell you to call 911, say okay. Once they say they are transferring you to a person, stop talking and end the call. Do not give your date of birth, insurance, or phone number. Do not ask to book, reschedule, cancel, add insurance, pay a bill, join a waitlist, get a text, or ask for the street, floor, or suite. Do not ask them to text you an appointment confirmation.

Caller traits

full_name
Bill Tanaka