context rot is the silent killer of AI output quality... i tested identical prompts with full context files in Gemini 3.1 Pro and Opus 4.6 Gemini's window is significantly larger... shouldn't that give it an edge? nope, both models hit the same wall at roughly the same point in the conversation output goes from sharp to generic to useless the window size is a distraction today, the real bottleneck is how you manage what goes inside it what matters now: - how token-efficient your context…