There's the real bug. The file is in git and on disk, sync reports it as "unchanged / already imported," but

gbrain get

says the page doesn't exist. So the index has a stale checkpoint — it's skipping the file based on git state without the page actually being in the DB. A

--full

didn't force a reimport because it's keying off content hash + checkpoint. Let me check the older tracked files to see how far back the gap goes, then force a real reind