ADR-0014 — Office documents and other files, kept as their original

Status: Accepted · Date: 2026-11-06 · Concerns the edge and the cloud · First step of the Archive (ROADMAP.md)

Context

The edge lists every file in a source and gives it a kind (engine.py), but until now it published only photos, RAW files and films. Office documents were counted and then left behind, and so was everything else: music, editing projects, archives, files of unknown type. A drive onboarded into a store (ADR-0012) was therefore only partly in the Mediathek.

The roadmap's Archive (previews, full text, cases, keep-until dates) is a larger piece of work. These files need a place in the Mediathek before that exists.

Decision

Two more sides, besides photos and films. Both the edge and the Library have them:

Side	Kinds	Extensions (edge, `engine.py`)
Office	`office`	pdf, doc(x), dot(x), rtf, odt, xls(x/m), ods, csv, ppt(x), pps(x), odp, txt, md, pages, numbers, key, vsd(x), one
Other	`audio`, `project`, `archive`, `other`	everything else, including svg and html

Two kinds are never published: sidecar (Takeout's JSON, already read into the photo it belongs to) and system (desktop.ini, Thumbs.db, .DS_Store).

The same process as photos. A publish of the new kinds goes through the same steps:

the dry run that says what would be published;
one upload per fingerprint (SHA-256);
the source's vault, with one vault per item (ADR-0011);
items left out by hand;
Move to vault.

The original only. These files have no renditions, so there is nothing under derived/. The Library shows a tile with the file type and the name. The viewer shows a card with a Download button, which uses originalUrl. originalUrl presigns with Content-Disposition: attachment, so an html or svg file is downloaded and never rendered in the Mediathek's origin.

Vault moves need no change. vaultMove lists derived/<from>/<id>/. For these files the list is empty, so steps 1 and 6 do nothing, and the row, albums and knowledge move as for a photo.

Over 5 GB is refused in the dry run. That is the single-PUT limit. Such a file is reported as too large instead of failing during the upload.

Consequences

An office document goes to its source's vault. Sensitive papers in a family source become visible to the whole family. Point such a source at a private vault before publishing it, or move the papers afterwards. The roadmap's question about a stricter vault for sensitive papers stays open.
Nothing is read from inside the files: there are no previews, no text and no search. Those belong to the Archive.
edgeMedia accepts the kinds office, project, archive and other. An edge older than this one never sends them. A cloud older than this one refuses them with a 400, so deploy the cloud first.