Start Decompositions

Which guidelines to decompose: the ones named, the ones with no requirements, or the stale ones.

Exactly one of the three, which the handler enforces with a 400 rather than a validator here — a 422's detail is a list of error objects and apiErrorDetail in the browser drops it for a generic message, so a rule expressed here is a rule nobody is ever told about. See _scope_of_the_request in backend/api/marketing/guidelines.py.

file_ids = None used to mean every marketing standard the company has, and a single button sent it: one press re-read documents whose libraries were already there, retiring each of those generations. A regeneration cannot carry a person's wording edits onto the generation that replaces it (content_edited_by_a_person exists to count what is lost), so that press destroyed work by design. That unnamed whole-library shape is gone. What replaced it are two named sets the server re-derives as the request lands — without_requirements and stale — so a caller cannot ask for a whole-library re-read of current libraries by leaving a field out, and neither flag can be widened into one without being renamed.

Naming files explicitly still re-reads whatever is named, whether or not it has a library: that is what the per-document Regenerate is for — a library built from a version the document has moved past, or a run that failed. stale is the same decision as that per-row button, asked of every standard that is in that state, still derived here rather than from a browser-assembled list.

The response names everything that was started and everything that was not, because the caller cannot otherwise tell a guideline that was skipped from one that silently failed.