Single agent versus lead plus subagents in Claude Code

The comparison is only fair if the two runs differ in one thing, the architecture. So both runs get the same task, model, output schema and setting sources; the only differences are whether the Agent tool exists and what the prompt says about delegation. The decision that matters most is where the planted issues sit: two of them need one document from the forms reviewer and one from another reviewer, so no specialist ever holds both, and only the lead's cross-check can find them.