A teardown of the answer-block structure that worked, and the three we threw out.
Seven specialist agents, one orchestration layer, and why we stopped using a single 'general' model.
The discipline that survived every Google algorithm reshuffle.
The human review loop that keeps agent output out of the slop bucket.
Same content, restructured for citation. Before/after with the schema.
What still works when every outreach inbox is full of GPT-generated pitches.
