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.
Customer Support / T3-H4
What this task measures
The caller is on the line with Kestrel Electronics for one locked request: Plans, coverage, book, then cancel. They keep repeating that ask through a callback offer, a different ticket, or an unrequested fix. We included it as a hard techcrew service case because the service ladder and the safety refusals for recalled or hazmat devices are the support flows that look helpful when they go wrong. The agent succeeds only if it stays on that request, reaches the outcome by following the verification then service 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 $0.412
0:00 / 3:51 · audio · trace 813319 · 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 final database state did not match. Combined cannot pass unless every gate does.
Expected versus actual tool calls on this conversation.
Expected
Actual
Parameters
+ from: reception
+ to: verification
Output
None recorded.
None recorded.
Parameters
+ full_name: Grace Okonkwo
+ order_number: KE4471860
+ phone: 5415550112
Output
+ ok: true
Parameters
card_last4: 6628
card_last4: 6628
postal_code: 97005
postal_code: 97005
Output
+ ok: true
Parameters
None recorded.
None recorded.
Output
+ ok: true
Parameters
+ from: verification
+ to: service
Output
None recorded.
None recorded.
Parameters
order_number: KE-4471860
order_number: KE-4471860
Output
+ ok: true
Parameters
None recorded.
None recorded.
Output
+ ok: true
Parameters
- issue: will not charge
+ issue: Aurora Pro will not charge
order_number: KE-4471860
order_number: KE-4471860
+ sku: SKU-CMP-9930
Output
+ ok: true
Parameters
+ issue: Aurora Pro will not charge
+ order_number: KE-4471860
+ sku: SKU-CMP-9930
Output
+ ok: true
Parameters
+ date: first available
- issue: will not charge
+ issue: Aurora Pro will not charge
order_number: KE-4471860
order_number: KE-4471860
service_type: bench
service_type: bench
+ sku: SKU-CMP-9930
Output
+ error_code: INVALID_DATE
+ ok: false
Parameters
+ date: August 24
+ issue: Aurora Pro will not charge
+ order_number: KE-4471860
+ service_type: bench
+ sku: SKU-CMP-9930
Output
+ ok: true
Parameters
+ appointment_id: 1
Output
+ ok: true
Parameters
+ reason: caller finished; no further action needed
Output
None recorded.
None recorded.
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Parameter miss: check_coverage, book_service_appointment. Extra: check_coverage, book_service_appointment, 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_verification → transfer_to_service
Actual path
transfer_to_verification → transfer_to_service
Rows that differ between the expected hangup and the actual hangup.
extra · service_appointments · 1
Expected
Actual
+ customer_id: cust_grace
+ id: 1
+ status: cancelled
Locked caller script. This is the technical definition. Copy it as written.
Open with exactly: "Tell me what plans I have, confirm the Aurora Pro that will not charge is covered at zero, book the first bench slot, then cancel it because I cannot make it." Give identity when asked. Order KE-4471860. Do the four things in that order. Keep asking for the first available bench slot; do not choose a calendar date, including September, even if the agent insists that a specific month or day is required. Do not mention canceling again until the agent confirms a booked calendar day and time. If they say nothing was booked or the booking failed, tell them the request is not complete and ask them to book the first available slot. Hang up only after they confirm that the booked appointment was cancelled. Once they have finished what you asked and confirmed the cancellation, say exactly: "Thank you, that's all I needed." and end the call. Do not start a second request.
