Wrote a small script to test how LLMs handle growing context sizes. It sends larger prompts step by step and measures slowdown or failure. Pretty rough but useful so far. Curious if others have ideas for improving it.