Battle-tested methodology for orchestrating many concurrent subagents to build or audit a large dataset — a 3-wave discovery→pull→synthesis pipeline with shared instruction templates, a strict no-delegation rule, file-on-disk completion checks, batching, incremental commits, and a recombine + row-integrity validation step. Use when a task is too large for one context (hundreds of companies/files/records), when you're fanning out research or extraction across batches, when subagents must write results to disk without clobbering each other, or when you need to detect and re-run subagents that stalled or self-delegated. Model-agnostic; complements task/agent tooling.
