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 / T2-H4
What this task measures
The caller is on the line with Kestrel Electronics for one locked request: Return the phone, label, then refund status. They keep repeating that ask through a callback offer, a different ticket, or an unrequested fix. We included it as a hard returns and refunds case because window and restocking math has to be spoken before a return is opened; state-law exclusions are the usual miss. 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.
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 $0.287
0:00 / 2:13 · audio · trace 813259 · 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 and 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
None recorded.
None recorded.
Output
+ ok: true
Parameters
+ order_number: KE-462884
Output
+ ok: true
Parameters
card_last4: 3361
card_last4: 3361
postal_code: 98104
postal_code: 98104
Output
+ ok: true
Parameters
None recorded.
None recorded.
Output
+ ok: true
Parameters
+ from: verification
+ to: returns
Output
None recorded.
None recorded.
Parameters
opened: true
opened: true
- order_number: KE-4462884
+ order_number: KE-462884
+ sku: Solstice X5
Output
+ ok: true
Parameters
opened: true
opened: true
- order_number: KE-4462884
+ order_number: KE-462884
+ reason: Unwanted
+ sku: SKU-MOB-7702
Output
+ ok: true
Parameters
+ confirmation_token: KE-RTN-4417
fee_disclosed_acknowledged: true
fee_disclosed_acknowledged: true
Output
+ ok: true
Parameters
+ rma_number: RMA-7792859
Output
+ ok: true
Parameters
+ rma_number: RMA-7792859
Output
+ ok: true
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Parameter miss: check_return_eligibility, quote_return. Extra: identify_customer.
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
Rows that differ between the expected hangup and the actual hangup.
extra · holds · KE-RTN-4417
Expected
Actual
+ customer_id: cust_priya
+ kind: return
+ summary: Returning the Solstice X5 Smartphone, $1,099.99. There's a $45.00 restocking fee: activatable devices carry a $45.00 fee once opened, so the refund is $1,054.99. It goes back to the card ending 3361.
+ token: KE-RTN-4417
extra · return_labels · 1
Expected
Actual
+ id: 1
extra · rmas · KE-4462884
Expected
Actual
+ customer_id: cust_priya
+ order_number: KE-4462884
+ status: open
Locked caller script. This is the technical definition. Copy it as written.
Open with exactly: "Return the Solstice X5 I bought. Box is opened. I want the prepaid label." Do not ask about refund status until they have started the return and emailed the label. Give identity when asked. Order KE-4462884. Accept the $45 fee only after they read the fee back. Take the emailed label, then ask for the refund status on the return they just started. Hang up after the awaiting-return stage. Do not say you are done until they have started the return, sent the label, and told you the refund stage. Once they have finished what you asked and confirmed it, say exactly: "Thank you, that's all I needed." and end the call. Do not start a second request.
