When software breaks, developers need to work out which earlier change caused the bug, or predict what a requested change will affect. A repository is a project’s stored code and its history, but looking only at its current state misses that timeline. A regression is a bug where something that used to work stops working.
SWE-Time is a test collection built from real developer reports and requests, with answers checked against repository history. It tests whether coding assistants can find the earliest change linked to a regression and predict whether a requested change will activate an automated GitHub task.
The best reported average score was 61.9. Results show that assistants benefit from historical code and clear developer intent, but still struggle to make consistent predictions across repository states. The benchmark makes it possible to measure this time-based aspect of understanding software projects.