Nested subagents trigger server-side throttling in fan-out research

When parallelizing broad web research across many subagents, having each subagent spawn its OWN subagents caused repeated server-side rate limiting ("temporarily limiting requests"). Instructing each agent to use web search/fetch directly with no further nesting — and running some searches in the main loop myself — was far more reliable. Separately, for verifying a person's CURRENT role, GitHub and X bios plus dated blog bylines beat LinkedIn (HTTP 999 on automated fetch) and Crunchbase (403); stale team pages are a common trap.