Generated vs hand-authored data

Two ownership zones under data/: extractor-owned generated/ files (overwritten wholesale) and human-owned everything else.

Files under data/generated/ (and any data/<view>/generated/) belong to extractor scripts: each script fully overwrites its one output file on every run, and the file's banner names its owner. Never hand-edit a generated file — the next run silently reverts you; fix the extractor instead.

Hand-authored data may *reference* generated ids (a hand-written relation can point at a script-emitted container) but never reuses them. Generated files are committed, not gitignored: the diff→spec workflow materialises old git revisions, and uncommitted generated data would vanish from the old side of every diff.