Every quarter ends with the same deliverable: the "what changed" memo. Not what happened — what changed: which narratives strengthened, which assumptions broke, where management language moved, what the smart commentary now worries about that it didn't in April. It's the highest-leverage document you write and the one that gets the least time, because it lands exactly when you're most buried.

The raw material is the quarter's accumulated tape — the interviews, calls and expert commentary you (ideally) collected as they happened. The memo is a synthesis problem: read ninety days of record, difference it against the prior quarter, and write it up with sources. Which is precisely the kind of work that shouldn't start from a blank page.

deep_context does the differencing: with the quarter's commentary imported and tagged in your library, scope questions across it — "what themes emerged this quarter that weren't present before? where did management language shift?" — and get answers cited to the exact chunks. The synthesis runs over what was actually said, not what you happen to remember from June.

chat does the shaping: take the grounded findings and draft the memo in your format — the sections your PM expects, the tone your desk uses, tight. It's the no-retrieval half of the workflow: pure writing over established facts. You review, adjust judgment calls, and ship a memo where every claim has a citation — in an afternoon instead of a lost weekend.