NemoDots

The problem: Agents act on old facts. In a multi-agent team, one agent often keeps acting on a fact that has since changed. The result is wasted work or the wrong fix applied to a live system. The idea: Receipts for every fact. Every observation gets a versioned receipt, and every task and fix records which receipts it used. When a source changes, dependent work is marked stale, and a final check blocks any fix built on stale receipts. The proof: A real app gets fixed. A real checkout service is secretly broken. The agents find and fix it with no hints. A real test order decides whether it worked, not an agent saying "fixed".