Your ORM and your analyst open the same file

Your API opens the database the fast way: better-sqlite3 against /hoody/databases/app.db. Native speed, no network hop. Your analyst opens the same database the convenient way: an HTTP query against the container's SQLite URL, from her laptop, mid-meeting. Same file, same moment, both writing.

This is normally the setup for corruption: SQLite locks at the file level, and mixing access paths is how you learn that. Under /hoody/databases/, the host's FUSE layer coordinates the writes, so the embedded library and the HTTP surface are two doors into one safe room. No replica to sync, no read endpoint to provision, no choosing between embedded and served. The ORM and the curl hit the same bytes.