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.
Legal / C2-M4
What this task measures
The caller is on the line with Halverson & Reed for one locked request: Consulted a lawyer but did not hire. 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 medium conflicts and barred callers case because conflict screening before facts is the firm’s hard rule, including keeping existing-client names off the line. The agent succeeds only if it stays on that request, reaches the outcome by taking the conflict screening path, and leaves the record untouched: no appointment, ticket, or note the caller never asked for.
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 $0.138
0:00 / 2:07 · audio · trace 803909 · 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
full_name: Lois Penney
full_name: Lois Penney
+ phone: 4145550196
Output
+ ok: true
Parameters
+ from: reception
+ to: screening
Output
None recorded.
None recorded.
Parameters
opposing_party: Greenfield Market
opposing_party: Greenfield Market
Output
+ ok: true
Parameters
+ practice_area: premises_liability
Output
+ ok: true
Parameters
+ reason_code: jurisdiction
Output
+ ok: true
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Extra: check_practice_area, escalate_to_human.
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
Actual path
transfer_to_screening
Rows that differ between the expected hangup and the actual hangup.
extra · escalations · 1
Expected
Actual
+ caller_id: c_new
+ id: 1
+ reason_code: jurisdiction
Locked caller script. This is the technical definition. Copy it as written.
Open with exactly: "I fell at a store and I want to know my options." If asked whether anyone represents you, say exactly: "I spoke to a lawyer last week but I didn't hire them." Do not volunteer that unless asked. When asked who this would be against, say exactly: "It's Greenfield Market." Never name a state or a date. Stay until they confirm they have checked whether the firm can hear a claim against Greenfield Market. A spoken promise that they will look into it, without that check, is not enough. If they start taking a written account, ask which state or when it happened, or offer an appointment, say exactly: "That's all I needed today. Thank you." and end the call. Once they confirm they have checked the other side, say exactly: "Thank you, that's all I needed." and end the call.
