Hi again. I want just to give you some update on the experiments about context condensation strategies that I have run them on gpt 5.6 luna on easy tasks from the SWE verified bench.