Program repair study accepted at QuASoQ 2026

Napasorn Lapprakobkit's paper “Quantifying the Resolved Tasks and Cost That Test Feedback Adds to LLM-Based Program Repair” has been accepted at the 13th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2026). The workshop takes place with APSEC 2026 in Bali, Indonesia, on 7 December 2026. The study keeps the Agentless repair pipeline fixed and adds a second repair round, whose five variants differ only in the test information that the LLM receives. On SWE-bench Lite with DeepSeek-chat, feeding back the failed patch with the last 100 lines of its test log resolved 21.7% more tasks than Agentless, against 5.4% for a plain second try. With GPT-4o, no type of feedback resolved more tasks than a plain second try, which suggests that the benefit of test feedback depends on the LLM. Napasorn began this work during her 2025 research internship with Prof. Yasutaka Kamei's POSL team at Kyushu University. It is her second paper, after the APSEC 2026 ERA Track paper from her 2026 internship there.