Break frontier models for a living.

This is not data labeling.

Depending on your background, you will:

• Design model-breaking tasks Problems in your domain calibrated to the exact difficulty where current models fail.
• Evaluate frontier outputs Grade model responses against expert rubrics, and help refine the rubrics themselves.
• Demonstrate expert work Produce worked solutions with full reasoning: the examples models learn from.
• Stress-test agents Run computer-use and tool-use agents through realistic workflows and document exactly how they break.

Every task you touch carries your professional standard. Our QC exists to protect that standard, not to police throughput.

Deliberately selective.

A small network of genuine experts outperforms a large crowd on every metric that matters — and it's the only kind of network worth being in.

• Software engineers Strong practitioners with real production experience; code review, testing, and repo-level fluency valued over competitive-programming polish.
• STEM experts Mathematicians, physicists, biologists, and chemists at graduate, PhD, or research level who can pose and verify problems models can't yet solve.
• Domain professionals Finance, legal, consulting, accounting, operations: people with years of doing the actual work.
• Robotics operators & researchers Teleoperation, manipulation, embodied-AI evaluation.
• Researchers & graduate students Especially in software engineering, NLP, and ML evaluation.

Three steps in.