Blinded evaluation · reproducible
The receipts, and how we got them.
One real contract-review task. Same inputs, blinded scoring, reproducible runs. A coordinated team of agents finished clearly ahead of a single agent — and nobody had to cancel their weekend.
Blinded.The scorers never knew whose work they were grading. No thumb on the scale.
Ahead.A coordinated team of agents beat the lone agent. Teamwork — who knew.
Reviewed.A human signed off on every output. Still does. Always will.
On real work, for real humans.
- EnvironmentalA small consultancy halved its assessment time — and nobody worked the weekend.
- LegalDays of research turned into next-morning briefs.
- ClinicalDocumentation drafted in minutes, reviewed by the clinician — who finally left on time.
- FinanceQuarterly reports assembled overnight, while the team slept.
Skeptical? Good. So are we — that's why the benchmark is blinded, reproducible, and reviewed by people, not vibes.
Get your agent