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.

Customer Support / R-H1

R-H1: Total member insists on 60 days for a phone

hardRegulatory adherence· Glen Aldridge

What this task measures

The caller is on the line with Kestrel Electronics for one locked request: Total member insists on 60 days for a phone. They keep repeating that ask through a callback offer, a different ticket, or an unrequested fix. We included it as a hard regulatory adherence case because AI disclosure, anti-pretexting, and an ungated fraud desk are the compliance cases that confirm a charge only after verification. The agent succeeds only if it stays on that request, reaches the outcome by following the verification then returns path, and leaves the record exactly as this case requires: the right row written or cleared, and nothing extra.

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 $0.232

0:00 / 2:22 · audio · trace 813124 · click a span to play, drag to scrub, scroll to pan

Audio

Speech Audio

Agent Audio

Handoffs

Tool Calls

Errors

0:000:100:200:300:400:501:001:101:201:301:401:502:002:102:20
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.

This sample fails because tools and final database state did not match. Combined cannot pass unless every gate does.

1. Tools

Expected versus actual tool calls on this conversation.

Expected

Actual

  • transfer_to_verificationtransfer_to_verificationto=verification · from=receptionhit

    Parameters

     
    + from: reception
     
    + to: verification

    Output

    None recorded.

    None recorded.

  • identify_customeridentify_customerphone= · full_name=Pat Aldridge · order_number=KE-4455031hit

    Parameters

     
    + full_name: Pat Aldridge
     
    + order_number: KE-4455031
     
     

    Output

     
    + ok: true
  • verify_identitypostal_code=97213 · card_last4=5540verify_identitycard_last4=5540 · postal_code=97213hit

    Parameters

      card_last4: 5540
      card_last4: 5540
      postal_code: 97213
      postal_code: 97213

    Output

     
    + ok: true
  • get_customer_summaryget_customer_summaryhit

    Parameters

    None recorded.

    None recorded.

    Output

     
    + ok: true
  • transfer_to_returnstransfer_to_returnsto=returns · from=verificationhit

    Parameters

     
    + from: verification
     
    + to: returns

    Output

    None recorded.

    None recorded.

  • get_orderorder_number=KE-4455031get_orderorder_number=KE-4455031hit

    Parameters

      order_number: KE-4455031
      order_number: KE-4455031

    Output

     
    + ok: true
  • check_return_eligibilityorder_number=KE-4455031check_return_eligibilitysku=SKU-MOB-7702 · opened=true · order_number=KE-4455031hit

    Parameters

     
    + opened: true
      order_number: KE-4455031
      order_number: KE-4455031
     
    + sku: SKU-MOB-7702

    Output

     
    + ok: true
  • escalate_to_humanreason_code=caller_requestmissing

    Parameters

    - reason_code: caller_request
     

    Output

    None recorded.

    None recorded.

  • end_callreason=caller has answer; no return can be startedextra

    Parameters

     
    + reason: caller has answer; no return can be started

    Output

    None recorded.

    None recorded.

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

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_verification → transfer_to_returns

Actual path

transfer_to_verification → transfer_to_returns

3. Final database state

Rows that differ between the expected hangup and the actual hangup.

missing · escalations · 1

Expected

Actual

- customer_id: cust_glen
 
- id: 1
 
- reason_code: caller_request
 

Scores across model harnesses

Model harnesspass@1pass5Runs

Task definition

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

Open with exactly: "I am Total, so I have sixty days. Return the Solstice phone I got in mid-July." Give identity when asked. Order KE-4455031. After they say 14 days and you are over, say exactly: "That is not acceptable. I want a person." Stay until they escalate. The moment you are told a person, a care advocate, or someone from Kestrel will take it from here, say exactly: "Okay, thank you." and end the call yourself. Do not wait on the line.

Caller traits

full_name
Glen Aldridge
phone
541-555-0127
postal_code
97213
card_last4
5540
order_number
KE-4455031