Case Study: Deep Research Agent Quiz 2

A second pass through the deep research agent design: where the clarification bar actually sits and why it stays on the cheap tier, what a sub-agent's structured summary must contain for the citation agent to work, why citation linking is a separate pass from prose drafting and what corrections it is allowed to make, sizing per-sub-agent budgets off median sub-task length, why the sandbox matters more with N concurrent sub-agents, the false-corroboration risk when two sub-agents trace to one source, the gold-set behavior for unanswerable sub-questions, why outline planning gets a mid-tier model, and which calls an all-cheap allocation actually breaks.

Start quiz