Editorial Workflow

Scholarly editions are built, not published in one stroke. A transcription or translation begins as a working draft, is checked by a second pair of eyes, and only then becomes something a reader should cite. Archetype models that discipline directly: every text attached to a manuscript image carries an explicit editorial status, moves between states through recorded transitions, and never loses the record of who changed what and when. For a research project, a library, or a museum, this is the difference between a database of assertions and a defensible scholarly edition — the provenance of each reading is part of the data, not a matter of institutional memory.

The lifecycle of a text

Each transcription or translation (an ImageText, bound to one manuscript image) holds a status drawn from four states: Draft, Review, Live, and Reviewed. The two publication states carry a deliberate distinction — Live marks a text released for public reading, while Reviewed records that it has additionally passed a formal editorial check. Both are publicly visible; Draft and Review are not. Public and anonymous API callers only ever receive Live and Reviewed texts, so unfinished work stays private to the editorial team while remaining fully editable. Staff see every state.

A text advances through an explicit transition endpoint rather than a bare status field, so no state change is silent. When an editor sends a draft for checking, they may nominate a reviewer; that assignment is cleared automatically once the text leaves the Review state, so the queue never misrepresents who is expected to look at a given entry. Assignment is advisory — it makes the queue honest about intent, but any backoffice (superuser) account may act on any text.

Archetype records transitions rather than enforcing a fixed graph of permitted moves. The platform tracks the full history of state changes; it does not currently forbid, for example, a jump straight from Draft to Live. Editorial policy about which moves are legitimate is left to the team, not hard-coded.

The review queue

Texts in the Review state surface in a dedicated queue, ordered oldest-pending-first so the long tail never gets lost. Each entry arrives with the context a reviewer needs in a single view: the last transition (who submitted it, when, and any accompanying note) and the nominated assignee. From there a reviewer can approve a text — promoting it to Live — or send it back to Draft, where a note explaining what needs changing is required. Both actions run through the same audited transition path, so an approval and a rejection leave equally complete records. For corpus-scale cleanup, the same transition, plus language tagging and deletion, can be applied to many texts at once as an all-or-nothing batch that still writes one transition record per text.

An attributed, dated audit trail

Two complementary logs sit underneath the workflow, and both are append-only.

• A status-transition log records every editorial move on a text — from_status, to_status, the acting user, a free-text note, and a timestamp. This is what powers the “who sent this for review, when, and why” view in the queue and the per-text history timeline.
• A general edit-event log, decoupled from any single table, records creations, updates, deletions, and moves to and from the trash across audited models (transcriptions, images, image annotations, and site settings and labels). It answers project-management questions such as “what changed since last week” independently of any one workflow.

Attribution survives the loss of a contributor’s account: actor references are nulled rather than cascade-deleted, so an event’s what and when remain even if the who is later removed. The acting user is captured at the point of the write, so attribution reflects who actually performed the change rather than who owns the record.

Because published texts can be exported as standalone TEI P5 documents (see Interoperability), the editorial state that gates public visibility also governs what leaves the platform as citable scholarship. For how these texts relate to images and regions, see the Data model; for how published work becomes findable, see Search & Browse.