Bring one queue. Keep the proof.
Pick one queue you already run — annotation, RLHF, red-team, or eval. We scope it, run it on your data in your tenant, and hand back a reviewed result you keep. First batch in weeks, not quarters.
One queue in. You bring the work you already run; we map the reviewers and the moment that has to hold up under audit.
A standardized review loop on your data, in your tenant. A first batch in units and a price band up front, so you know what you are committing to before you send the form.
A reviewed result you keep. Every reviewed case, override, and regression carries who created it, who reviewed it, and under what rights — a record that survives an examiner, never pooled.
Scope the pilot.
Tell us the one queue you want to bring, where it can fail, and what proof has to survive an examiner. We reply in one business day with a scoped pilot, the deliverable defined, and the terms to expand or stop. Not a generic sales call.
Bring your hardest case.
Your largest source was absorbed by one of the labs it served. A competitor lost four terabytes, including who its workers were. The EU AI Act provenance clock starts August 2026. A pilot is how you prove a neutral, review-ready second source on your own work.