[TriangleEdge]

I'd like to see <super-long-context> vs <mediocre-context-but-really-good-at-summarizing>, then it is fed parts of a super long context in chunks. On each iteration, you add the previous summary.