AI agents can repeat investigations when they cannot see each other's findings. This proposed architecture uses shared work records and evidence summaries to help readers find related work and assess its limits.