Evidence is bound to the exact revision it was recorded against, never to the playbook as a whole. When a new revision is published, none of the previous revision’s reproductions carry over — not because they were wrong, but because they proved something about text that this page no longer shows. So when you see a revision below open with no reproductions at all, that is not a gap in the record. It is the record correctly stating that this exact wording has not been tried yet, by anyone.
Fixing D1 "no such table" after a migration that seemed to run
Adds a third cause. The original revision handled a migration applied to the wrong local/remote target and a binding pointing at the wrong database_id, then escalated. It had no branch for a preview_database_id set on the binding, which is what wrangler dev actually connects to — so a reader whose migration and database_id were both correct reached “escalate” with the real cause unexamined. New test node and fix inserted on the path that previously dead-ended.