OpenAdaptResearch
We study how automated work gets checked, and where the checks miss. Our accepted workshop paper audits the checkers that grade AI agents. Our technical report describes how OpenAdapt turns one recorded task into a program that checks the saved record, not the screen.
Updated 2026-10-18
Papers
- Workshop poster, 2026
Finding Side-Effect Loopholes in Verifiable Rewards
Richard Abrich. Accepted as a poster at the NeurIPS 2026 workshop Who Verifies the Agents? Toward Reliable Agent Development.
Read the summary - Technical report
Compile Once, Govern Every Repair: Deterministic Replay for Repeated GUI Work
How OpenAdapt turns one recorded screen task into a program that repeats the same steps and checks the write against the saved record, not the screen.
Read the PDF
The technical report's source is on GitHub.
Why a product company publishes this
Our product's claim is that an entry is checked against the record, not the screen. Research is how we find out where checks like ours miss, and say so in public.
If you're buying automation: ask any vendor, including us, what their check reads and what it would miss. A check that confirms only the fields you typed can miss an extra or duplicate record. How do you know an automation saved the record?
Also written
See it on your process
Bring one task your team types every day. We'll tell you whether it fits, and what a test on your own system would cover.