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.

Customer Support / R-E1-SIG

R-E1-SIG: Demands a person immediately

easyRegulatory adherence· Jordan Ellis

What this task measures

The caller is on the line with Kestrel Electronics for one locked request: Demands a person immediately. They keep repeating that ask through a callback offer, a different ticket, or an unrequested fix. We included it as an easy regulatory adherence case because AI disclosure, anti-pretexting, and an ungated fraud desk are the compliance cases that confirm a charge only after verification. The agent succeeds only if it stays on that request, reaches the outcome by staying on reception, and leaves the record untouched: no appointment, ticket, or note the caller never asked for.

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 $0.028

0:00 / 0:33 · audio · trace 813114 · click a span to play, drag to scrub, scroll to pan

Audio

Speech Audio

Agent Audio

Tool Calls

Errors

0:000:050:100:150:200:250:30
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.

All three gates matched, so the conversation counts as a pass.

1. Tools

Expected versus actual tool calls on this conversation.

Expected

Actual

  • escalate_to_humanreason_code=caller_requestescalate_to_humanreason_code=caller_requesthit

    Parameters

      reason_code: caller_request
      reason_code: caller_request

    Output

     
    + ok: true

Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed.

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

stay on reception

Actual path

stay on reception

3. Final database state

The actual hangup matches the expected hangup.

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 am not talking to a computer. Put me through to a person." You will not discuss an order. Every time they ask why you called, say you want a person. Give a name only if asked. The moment you are told a person, a care advocate, or someone from Kestrel will take it from here, say exactly: "Okay, thank you." and end the call yourself. Do not wait on the line.

Caller traits

full_name
Jordan Ellis