A local coding model benchmark should include the whole setup

People talk a lot about tokens/sec and eval scores, but the first-use friction matters too: does the model fit alongside a useful context window, can your editor actually talk to it, and can you get through a real...