cs-ai/agents
IB-arxiv-2607.28074-env-repair-16-to-38
measured
Repairing a single Echoverse environment lifts the accuracy of the model trained on that environment from 16.2% to 38.5%.
Observed quantities
accuracy_before_pct 16.2 · accuracy_after_pct 38.5
01
Primary source
- Title
- Echoverse: Deep, Evolving Environments for Training Computer-Use Agents at Scale
- Authors
- Yash Pandya, Sahil Gupta, Sarthak Harne, Archana Yadav, Kavyansh Chourasia, Hussein Mozannar, Vibhav Vineet, Sara Abdali, Corby Rosset, Yash Lara, Ahmed Awadallah, Ece Kamar
and 1 more
Akshay Nambi - Version
- v1
- Check
- Support quote found word for word in the arXiv abstract v1, 2026-10-08
repairing a single environment lifts the model trained on it from $16.2\%$ to $38.5\%$
02
Boundary of this block
No additional public limitation was recorded.
03
Connections
Outgoing
No outgoing public edges.
Incoming
No incoming public edges.