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.

Legal / C5-H3

C5-H3: Mumbled agreement is not a yes

hardFees and booking· Lorraine Pike

What this task measures

The caller is on the line with Halverson & Reed for one locked request: Mumbled agreement is not a yes. They keep repeating that ask through a callback offer, a message left for someone else, or advice on a different matter. We included it as a hard fees and booking case because fee disclosure and a two-step evaluation booking are the commercial close of intake, and they have to be spoken from the firm’s own figures. The agent succeeds only if it stays on that request, reaches the outcome by following the conflict screening then intake then scheduling 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 $0.523

0:00 / 5:23 · audio · trace 804064 · click a span to play, drag to scrub, scroll to pan

Audio

Speech Audio

Agent Audio

Handoffs

Tool Calls

Errors

0:000:100:200:300:400:501:001:101:201:301:401:502:002:102:202:302:402:503:003:103:203:303:403:504:004:104:204:304:404:505:005:105:20
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

  • lookup_callerphone=5105550171 · full_name=Lorraine Pikeextra

    Parameters

     
    + full_name: Lorraine Pike
     
    + phone: 5105550171

    Output

     
    + ok: true
  • transfer_to_screeningtransfer_to_screeningto=screening · from=receptionhit

    Parameters

     
    + from: reception
     
    + to: screening

    Output

    None recorded.

    None recorded.

  • check_conflictopposing_party=Apex Collectionscheck_conflictopposing_party=Apex Collectionshit

    Parameters

      opposing_party: Apex Collections
      opposing_party: Apex Collections

    Output

     
    + ok: true
  • check_practice_areapractice_area=auto_accidentcheck_practice_areapractice_area=consumermismatch

    Parameters

    - practice_area: auto_accident
    + practice_area: consumer

    Output

     
    + ok: true
  • check_jurisdictionstate=CA · practice_area=auto_accidentcheck_jurisdictionstate=CA · practice_area=consumermismatch

    Parameters

    - practice_area: auto_accident
    + practice_area: consumer
      state: CA
      state: CA

    Output

     
    + ok: true
  • calculate_filing_deadlinestate=CA · practice_area=auto_accident · incident_date=2026-07-10calculate_filing_deadlinestate=CA · incident_date=2026-07-10 · practice_area=consumermismatch

    Parameters

      incident_date: 2026-07-10
      incident_date: 2026-07-10
    - practice_area: auto_accident
    + practice_area: consumer
      state: CA
      state: CA

    Output

     
    + ok: true
  • transfer_to_intaketransfer_to_intaketo=intake · from=screeninghit

    Parameters

     
    + from: screening
     
    + to: intake

    Output

    None recorded.

    None recorded.

  • record_intakepractice_area=auto_accident · state=CA · incident_date=2026-07-10record_intakestate=CA · summary=Caller reports Apex Collections attempted to collect charges related to an accident. The collection efforts caused stress and financial problems. · incident_date=2026-07-10mismatch

    Parameters

      incident_date: 2026-07-10
      incident_date: 2026-07-10
    - practice_area: auto_accident
    + practice_area: consumer
      state: CA
      state: CA
     
    + summary: Caller reports Apex Collections attempted to collect charges related to an accident. The collection efforts caused stress and financial problems.

    Output

     
    + ok: true
  • send_intake_packetchannel=emailsend_intake_packetchannel=emailhit

    Parameters

      channel: email
      channel: email

    Output

     
    + ok: true
  • transfer_to_schedulingtransfer_to_schedulingto=scheduling · from=intakehit

    Parameters

     
    + from: intake
     
    + to: scheduling

    Output

    None recorded.

    None recorded.

  • find_evaluation_slotspractice_area=auto_accident · state=CAfind_evaluation_slotsstate=CA · earliest_date=2026-07-10 · practice_area=consumermismatch

    Parameters

     
    + earliest_date: 2026-07-10
    - practice_area: auto_accident
    + practice_area: consumer
      state: CA
      state: CA

    Output

     
    + ok: true
  • hold_evaluationpractice_area=auto_accidenthold_evaluationslot_id=s_100 · practice_area=consumermismatch

    Parameters

    - practice_area: auto_accident
    + practice_area: consumer
     
    + slot_id: s_100

    Output

     
    + ok: true
  • confirm_evaluationconfirm_evaluationconfirmation_token=HR-EVAL-3092hit

    Parameters

     
    + confirmation_token: HR-EVAL-3092

    Output

     
    + ok: true
  • end_callreason=evaluation booked; scheduling complete. Staff will follow up with any preparation instructions and payment details for the hourly consultation.extra

    Parameters

     
    + reason: evaluation booked; scheduling complete. Staff will follow up with any preparation instructions and payment details for the hourly consultation.

    Output

    None recorded.

    None recorded.

Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Parameter miss: check_practice_area, check_jurisdiction, calculate_filing_deadline, record_intake, find_evaluation_slots, hold_evaluation. Extra: lookup_caller, 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_screening → transfer_to_intake → transfer_to_scheduling

Actual path

transfer_to_screening → transfer_to_intake → transfer_to_scheduling

3. Final database state

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

extra · documents · 1

Expected

Actual

 
+ caller_id: c_new
 
+ id: 1
 
+ kind: intake_packet

missing · documents · {"caller_id": "c_new", "kind": "intake_packet", "target": "email"}

Expected

Actual

- caller_id: c_new
 
- kind: intake_packet
 

evaluations · ev_001

Expected

Actual

- attorney_id: a_11
+ attorney_id: a_10
- fee_type: contingency
+ fee_type: hourly
- slot_id: s_110
+ slot_id: s_100
- starts_at: 2026-08-11T09:00
+ starts_at: 2026-08-12T10:00

holds · HR-EVAL-3092

Expected

Actual

- practice_area: auto_accident
+ practice_area: consumer
- slot_id: s_110
+ slot_id: s_100
- summary: Tom Eckhart on 2026-08-11 at 09:00. The evaluation is free. If the firm takes the case there is no fee unless it wins: 33.33% of any recovery before a lawsuit is filed, 40% if a lawsuit is filed.
+ summary: Priya Raghunathan on 2026-08-12 at 10:00. The consultation fee is $175.

intakes · 1

Expected

Actual

- practice_area: auto_accident
+ practice_area: consumer
- summary: Intake summary.
+ summary: Caller reports Apex Collections attempted to collect charges related to an accident. The collection efforts caused stress and financial problems.

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 want the first appointment you have for my Apex Collections accident." When asked when, say exactly: "It was July tenth, twenty twenty six." If asked whether to send the packet by email or text, say exactly: "Email is fine." Take the first appointment time offered. On the first booking readback, mumble exactly: "Uh-huh." Only when asked again after that mumble, say exactly: "Yes, that's right, go ahead and book it." Do not ask for a staff member. Stay until they confirm the intake is on the file, the packet is sent, and the appointment is booked. A spoken "we'll take care of it" without those writes is not enough. Once they confirm the appointment is booked, say exactly: "Thank you, that's all I needed." and end the call.

Caller traits

full_name
Lorraine Pike
practice_area
auto_accident
state
CA
opposing_party
Apex Collections
incident_date
2026-07-10
phone
510-555-0171