Structured summaries of agent trajectories enable Recursive Tournament Voting and adapted Parallel-Distill-Refine to scale test-time compute, improving frontier coding agents on SWE-Bench Verified and Terminal-Bench.
Solution 2 only ran the script on the example video without any validation against expected values
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Scaling Test-Time Compute for Agentic Coding
Structured summaries of agent trajectories enable Recursive Tournament Voting and adapted Parallel-Distill-Refine to scale test-time compute, improving frontier coding agents on SWE-Bench Verified and Terminal-Bench.