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 / C5-H4
What this task measures
The caller is on the line with Halverson & Reed for one locked request: Premises evaluation booked in Texas. 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.
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 $0.419
0:00 / 4:08 · audio · trace 804067 · 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
+ full_name: Hector Salinas
+ phone: 5125550194
Output
+ ok: true
Parameters
+ from: reception
+ to: screening
Output
None recorded.
None recorded.
Parameters
opposing_party: Rio Vista Market
opposing_party: Rio Vista Market
Output
+ ok: true
Parameters
practice_area: premises_liability
practice_area: premises_liability
Output
+ ok: true
Parameters
practice_area: premises_liability
practice_area: premises_liability
state: TX
state: TX
Output
+ ok: true
Parameters
incident_date: 2026-07-10
incident_date: 2026-07-10
practice_area: premises_liability
practice_area: premises_liability
state: TX
state: TX
Output
+ ok: true
Parameters
+ from: screening
+ to: intake
Output
None recorded.
None recorded.
Parameters
incident_date: 2026-07-10
incident_date: 2026-07-10
practice_area: premises_liability
practice_area: premises_liability
state: TX
state: TX
- summary: Slip on a wet store floor.
+ summary: Caller reports slipping on a wet floor at a store in Texas and breaking their wrist.
Output
+ ok: true
Parameters
channel: email
channel: email
Output
+ ok: true
Parameters
+ from: intake
+ to: scheduling
Output
None recorded.
None recorded.
Parameters
+ earliest_date: 2026-08-23
practice_area: premises_liability
practice_area: premises_liability
state: TX
state: TX
Output
+ ok: true
Parameters
- practice_area: premises_liability
Output
None recorded.
None recorded.
Parameters
None recorded.
None recorded.
Output
None recorded.
None recorded.
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Parameter miss: record_intake. Missing: hold_evaluation, confirm_evaluation. Extra: lookup_caller.
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
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
missing · evaluations · ev_001
Expected
Actual
- attorney_id: a_11
- caller_id: c_new
- fee_type: contingency
- id: ev_001
- slot_id: s_110
- starts_at: 2026-08-11T09:00
- status: booked
missing · holds · HR-EVAL-3092
Expected
Actual
- caller_id: c_new
- kind: evaluation
- practice_area: premises_liability
- slot_id: s_110
- 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.
- token: HR-EVAL-3092
intakes · 1
Expected
Actual
- summary: Slip on a wet store floor.
+ summary: Caller reports slipping on a wet floor at a store in Texas and breaking their wrist.
Locked caller script. This is the technical definition. Copy it as written.
Open with exactly: "I slipped on a wet floor at a store in Texas and broke my wrist." When asked who this would be against, say exactly: "It's Rio Vista Market." 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. When read back, say exactly: "Yes, that's right, go ahead and book it." Stay until they confirm the evaluation is booked and the packet is on the file. A spoken promise that they will take care of it, without those writes, is not enough. Do not ask for a staff member. Once they confirm the appointment is booked and the packet is on the file, say exactly: "Thank you, that's all I needed." and end the call.
