Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-08-15 · source-backed
Characterizing 28 repair approaches on SWE-bench Verified, the median produces 121.78% more total changes, 80.91% more net changes, and 43.99% higher cyclomatic complexity than the human fix. RECAP separates minimization from generation: run your pipeline unchanged, then apply a post-hoc refinement pass that strips redundant edits while re-checking tests. Average total changes fell from +242.14% to +4.24% with resolved instances held or improved. arXiv 2608.13292 Baselines that shrank patches by prompting lost 49 to 217 resolved instances doing it. This is the cleanest actionable result in today's papers.
Each link below shares sources, entities, or timing with this story.
arXiv 2608.13292 characterized 28 APR approaches on SWE-bench Verified: median 121.78% more total changes, 80.91% more net changes, 43.99% higher cyclomatic complexity than the developer patch, even when correct. The verbosity is rooted in capability-oriented design and resist...
The Agent Lightning result (arXiv 2608.17528) is a 9B model gaining 14.6 points from 3,500 lines of training code and modest compute. The point isn't that a 9B beats anything, it's the cost curve: 6K examples is a dataset a small team can build. Task-specific agentic RL on an...
A June paper shows source code is the dominant token sink for state-in-context SWE agents, and minification (stripping non-essential lexical elements, preserving semantics) cuts average input tokens 42% for only a 12-point SWE-bench Verified drop. Public on GitHub. For agents...
The study extracted 130 clean atomic state transitions from 707 real issues in SWE-bench Lite and Verified. Plain RAG scored 0.57-0.59 answer accuracy; an LLM reranker didn't help and added latency, about 18 seconds against 2.1. A (subject, relation, object) supersession memor...
Agent Lightning v1.0 (arXiv 2608.17528) inverts the standard agentic RL architecture, and the inversion is the whole point. Normally the training engine owns the environment loop. It drives the agent, collects trajectories, computes rewards. Which means your training setup and...
RETRACE has a verifier infer what problem the patch appears to solve using only the patch and trajectory, then compares that inference against the real issue. Training-free, lifted Pass@1 by 7.0% and 3.6% on mini-SWE-agent over SWE-bench Verified. The information-hiding trick...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.