6 tasks
Extract
How accurately does the system turn documents into data?
EnviroBench. The benchmark for AI in environmental work.
Statvis built EnviroBench and runs it: lab tables, scanned reports, site plans and project files, scored case by case against a fixed answer key. Every change to Statvis is scored on it before it ships.
What it measures. Six categories, scored 0 to 100.
Seventeen tasks, from reading a single table cell to working from a site’s whole record. Answers are never shown to the system under test, every case is scored on its own, and the real-world test sets stay private, so they can’t be trained on.
1 task
Verify
Does the system know when it might be wrong?
1 task
Locate
Can the system place things on site plans?
3 tasks
Research
Does the system find the right evidence, and see where the record conflicts or falls short?
4 tasks
Legal
Does the system read legal instruments without overstating them?
2 tasks
Site
Can the system work from a site’s whole document record?
Take control of your contaminated sites.
Bring us one site's documents. We'll show you what's in them.