The L1 single-operation track of SOLExec Bench has been confirmed as first place at a score of 0.7608, on Databricks' entry, with kernels generated by KDA. The previous best was 0.7584, by Recursive AI.
The margin is 0.0024, which is worth being plain about: this is not a result where an agent loop found something nobody had thought of. It is a result where a track that many people had already worked over moved again, by a small amount, because the search was run for longer and more carefully than a person would sit through.
That is most of what these applications are for. The dramatic wins make better posts; the narrow ones on picked-over tracks are the better evidence that the method is doing something repeatable.
The result is currently private on the board and can be made public at any time.