Interesting approach. The key unlock for multi-agent reliability is making memory writes auditable and reversible, so transient observations don’t silently become facts for downstream agents.