If it fits in the context window, feed it all in.

Mostly wrong. Million token windows made people lazy. More context isn't more understanding. Past a point it's a slower, pricier way to get a worse answer.

Context windows got huge, and a habit came with them. Why summarize, why retrieve, why choose, when you can paste the whole thing in and let the model sort it out. It feels thorough. It is usually a mistake.