Help articles
Review
Curating the Batch Roll-up
Merge duplicates, edit labels, remove wrong extractions, and select access points for sends when your workspace holds a connection.
On this page
Section links appear after the article loads.
Curating the Batch Roll-up
The roll-up gathers every entity and subject heading across the batch into one curation surface. One row per identity, with its authority chips, its document receipts, and the verbs to judge it. It is derived, never stored as a separate copy. It recomputes from your documents plus your saved curation, so item-level fixes flow in automatically.
Selecting access points
Selection chooses access points for sends, and it is opt-in. Nothing is ticked by default, and access-point sends deliver exactly what the saved curation shows. Ticking never removes anything. Item records and file exports carry every mention regardless. Checkboxes and per-section Select all and Clear appear when your workspace holds an ArchivesSpace connection, yours or a teammate's. Without one, the roll-up offers the verbs that shape your exports, merges, labels, roles, and removal, and file exports remain a complete delivery on their own. Subject headings come first, then agents and entities grouped by reconciliation status, with broader geography collapsed by default.
Save curation persists everything on the surface at once, and every save keeps an undo history you can restore from.
Merging duplicates
When the same identity appears under two spellings, "Casals, Pablo" and "Pau Casals", merge the variant into the canonical row. The merge picker ranks likely targets by name overlap, with authority-linked rows boosted. Merges are reversible (unmerge from the chip under the target row), and the relationship graph re-routes its edges to the canonical row immediately.
Merged spellings aren't discarded. When an agent is created in your repository, they ride along as variant name forms, so the record knows every way the archive wrote the name.
Subject headings offer merge and label editing only while they are unreconciled. Once a heading carries a FAST or LCSH link, the vocabulary supplies the label and identical headings converge on their own, so those rows offer selection and chip removal instead. An unreconciled heading can still be merged into a linked one to adopt the authorized form. Any heading that is wrong for the material, linked or not, can be removed from every record at once; the confirm names the document count, exports and sends stop carrying it, and for a few moments after, an Undo on the toast puts it back.
Labels, roles, and authorized forms
Edit a row's label to the form your catalog should carry, or fetch the authorized form (Library of Congress first, Wikidata fallback) with one click. Label edits are curation, not data edits. The documents keep their own spellings, and the pre-edit form becomes a variant.
Roles, creator versus subject, are suggested from document evidence (who wrote, who is written about) and can be overridden per row. A partial creator share gets an amber hint asking you to confirm the role. When the role truly differs by document, open the row's document list and set creator or subject per document. The distinction is stored on the document itself, beside the role the pipeline extracted, and flows back into the roll-up, item sends, and exports on the next derive.
Repository reconciliation
With ArchivesSpace connected, the roll-up checks whether each agent already exists in your repository
- Linked · authority identifiers match; sends reuse the repository's record automatically
- Suggested · name forms match (including variants); you confirm or dismiss, and can adopt the repository's authorized form in one click
- Conflict · the same name carries different identifiers on each side; amber-flagged for your judgment, never auto-picked
Relationship rows get the same awareness. When both endpoints match repository agents that are already related, the row says so, so you don't re-send what the repository already knows.
Confirmed identities across batches
Confirming "same identity" is remembered across your whole workspace, per repository. The next batch that mentions that person resolves them automatically, no re-asking, and your confirmed spellings are written onto the repository agent as variant name forms, so its own search learns them too.
Confirmations can be retracted at any time; retraction applies workspace-wide and takes effect on future sends. Variant name forms already written to the agent remain.
Relationships
The Relationships section gathers connections between agents. Asserted in the documents themselves, or derived deterministically from roles. Correspondence pairs from letters, interview pairs, agreement parties, letterhead affiliations. Asserted relationships are ready to send unless you exclude them; derived relationships send only when you confirm them; everything else stays visible but never sends. Without a workspace connection the section shows one flat list, relationships stay visible with their evidence, and inclusion appears once the workspace holds a connection.
Every edge carries its evidence quote and its document receipts. No numeric scores are invented for relationships. The evidence is the argument.
Where to go next
- §7 Delivering & the ArchivesSpace Integration · What happens to your curation at delivery time
- §5 Reviewing in the Workbench · Item-level review and editing, and the item-versus-batch boundary