The authors defined two runtime potentials: navigation potential measures exposure to task-relevant files, while verification potential measures alignment between the repository’s current test state and the desired outcome.
Runtime milestones improve credit assignment for long-horizon software engineering agents
SWE-MILE uses repository navigation and test-state changes to supervise individual actions while verification runs asynchronously in a sandbox.
Chinese Tech
Chaoqun Cui · Hao Zhou · Meiqi Chen · Fandong Meng · Wenji Mao
MAIS, Institute of Automation, Chinese Academy of Sciences · School of Artificial Intelligence, University of Chinese Academy of Sciences · Tencent Inc., China
Research Digest··2 min read
Cui et al.
Why this paper
From MAIS, Institute of Automation, Chinese Academy of Sciences and 2 others
In one line
Asynchronous potential-induced milestone credit assignment improves long-horizon software engineering agent performance by deriving process supervision from workflow runtime signals.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§