The paper surfaces a pre-PR coordination layer that existing PR-history analysis cannot see, and provides a concrete substrate and mining toolkit that reduce redundant multi-agent work from 78% to 0% — directly addressing why autonomous agents' PRs are accepted less often despite being produced faster.