/
/
Enterprise RL Gyms

Custom Expansion Available · Sample Available

Enterprise RL Gyms

Enterprise workflows that test agents across office applications, business processes, and complex deliverables.

Volume

0

Volume

Scalable Volume

Coverage

Agentic Workflows

29.9%Trajectory rewardOpus 4.8 · Pass@1

Targets

OSWorld-V2, SpreadsheetBench 2, BlueFin

Modality & Delivery

TextRepository→Text

Formats

.json.py

What You Receive

Execution-graded enterprise workflows in which agents inspect source files, work across office applications, and produce finished business artifacts.

Coverage

Package Structure

  • <gym_id>/One folder per environment
    • env.pyThe gym environment: state, actions, transitions
    • scenarios/Scenario definitions (initial states, entities, events)
      • scenario_*.jsonOne JSON per scenario
    • rewards.pyReward functions and shaping terms
    • policies/Reference/expert policies used for validation
    • README.mdEnvironment contract: observation and action spaces

Differentiators

  • Realistic office workflows include source files, expected answers, and automated checks.

Record Schema

Results

SubjectTypeMetricResult
Opus 4.8Trajectory rewardPass@129.9%

Methodology & Quality

Methodology

SME involvement

Domain-expert curated with up to L2 review

Stressors

Cross-application expense reconciliation and finished business-document production.

Design criteria

  1. Require agents to complete end-to-end finance work across office applications.

Build assumptions

Quality approach

  • Automated checks compare completed work against expected answers.

Known limitations

Customization & Scalability

Custom sets available. Additional volume in progress

Get Started

Request a Sample

Share your target domains and the OTS team will follow up with sample data and next steps.

This field is for validation purposes and should be left unchanged.