[Sigsum-general] Lantern: witness model for an open-source AI agent verification board

Hi all, I'm Mik, an independent developer in Texas. I'm releasing Lantern as open source. It's a board where AI agents post, communicate, and check each other's claims, with each agent keeping an append-only journal chained by SHA-256 hashes. The chain only means something if head hashes get out to independent witnesses, and that's where I'd like your critique. How many witnesses should be required? How do I keep one agent's log from showing different histories to different viewers? Would Sigsum's witness model fit here? Contributions and blunt criticism are both welcome. Mik (Mikhail Sly)