Help articles
Review
Reviewing in the Workbench
Triage by confidence, edit titles and descriptions, fix entities at the source, and act on sensitivity flags.
On this page
Section links appear after the article loads.
Reviewing in the Workbench
The workbench is where you judge what Acervo produced, correct what needs correcting, and decide what goes forward. Titles and descriptions are edited in place, wrong authority identifiers are fixed at the source, and sensitivity flags carry real consequences into delivery.
The rule of thumb. Identity-shaped judgments happen at the roll-up, one row per person or subject, the verbs it needs on the row. Document-shaped facts, a title, a description, a per-document role, a digital identifier, happen on the document. That is the whole seam.
If you haven't read it yet, §4 Processing & Outputs covers what each output is and how it was generated. This section walks through the review workflow; §6 Curating the Batch Roll-up covers batch-level curation.
The three tabs
Every batch opens to three tabs.
- Review · the batch-wide view. Signal tiles, the About-this-batch strip (housed in, accession, digital access), the access-points roll-up, and the low-confidence triage table
- Documents · every document record, for item-level inspection and editing
- Deliver · exports, direct sends, readiness checks, and narrative drafts (see §7 Delivering & the ArchivesSpace Integration)
When a scan holds more than one unit
Acervo reads a scan as one document, but when it sees an envelope, a letter with its enclosures, or several items in a single file, the classification card says how many units it found and lists each one with its page range and page notes. In a compound scan, an envelope or a letter with enclosures, the primary unit stays the document and each other unit carries a Split out verb that lifts its pages into their own record, where you name the document type and set its rights. When the scan is a file of distinct documents filed together, there is no primary unit. Acervo describes the scan at the file level, its type reads File, and every unit can be split into its own record. Splitting costs nothing, since segmentation and the read were already part of the original document.
A split-out record begins as a stub, with its text already in hand and its own place in the batch. Describe this unit runs the full read over that stored text so the unit gains its own entities, metadata, and authorities. Describing a unit charges per page at the batch's model rate; segmentation and the original read were already paid on the source document, so they are not charged again.
Triaging by confidence
You don't have to review every claim with equal scrutiny. Low-confidence instances are collected in the triage table on the Review tab, worst first, so you start where attention is actually needed. Each entity's confidence is built from two questions, Source Text (how clearly the document supports it) and Visual Legibility (how readable the underlying image is).
Subject headings carry a score and a rationale on hover; FAST and LCSH chips link to the authoritative entry so you can verify a heading against the controlling vocabulary in one click. Authority matches on entities are binary. Either a match was committed (with a link you can check) or it wasn't, and absence of a link means the matching pipeline could not find a candidate it was confident about. That's a signal for cataloger judgment, not an error.
Editing titles and descriptions
Each completed document card on the Documents tab carries an editable title, description, and document date. Edits write through to the stored record, so they ripple into every export, every send, and the batch's derived name. Acervo's original value is kept alongside your edit. An "edited" badge marks the change, and one click restores the original.
The visible box edits the title and description that exports and sends use. The document date field beneath them edits the one date that dates the record in every export and, when ArchivesSpace is connected, becomes the creation date on the sent record. Titles are prose and carry no date of their own. The card composes its displayed title from the title and the document date, exactly the way the delivered record will read.
The card shows a document's title in the document's own language and its description in your institution's language of description, set on the institution profile. When a stored variant is missing, it falls back to English. This chooses which stored variant you see and edit; it is separate from your interface language.
Fixing entities at the source
Data corrections happen at the item level, where the stored record lives. Removing a wrong authority chip edits the stored data across every affected document in the batch, after one confirmation, and adding a missing authority id works the same way, one identity, written everywhere it appears. A wrongly merged identity (two people collapsed because they shared a wrong identifier) separates on its own the next time the roll-up derives.
On the document's entity table, every change saves as you make it. There is no save button to remember. Edit a row, add an entity or a subject heading, or remove a wrong authority chip from that one document's record, and the stored record updates at once. Removing an extracted entity is reversible, the row moves to a Removed from this record list with one-click restore, and Restore original extraction brings back the pipeline's untouched output whenever you need to start over. Authority links appear as named chips, FAST, LCNAF, VIAF, Wikidata, GeoNames, each opening the authority's own record, and every row opens an evidence drill with the written forms, pages, and confidence receipts behind the claim.
Label spellings, roles, and merges are batch-level curation, not item edits. They live in the roll-up. See §6 Curating the Batch Roll-up for the boundary. Fix wrong data at the item; curate identity, and selection when your workspace holds a connection, at the batch.
Sensitivity flags
Processing flags entities and values that look like private financial details, personal identifiers, medical information, or harmful language. Flagged documents show a review notice on their card so you check the material before it goes anywhere public.
A flagged record is held back from publication until a person clears the flag. When a flag is a false positive, clearing it takes one click and a confirmation. Publication is always a human decision.
Finding past batches
Past work is organized as batches, in the sidebar's Recent list and on the History page. Outputs stay available there for 90 days.
The History page has a search box. Type a batch name, a document title, a filename, or a collection name and the list narrows to the batches that match, with the matching documents shown inside each. Accents do not matter; Pena finds Peña.
See §10 Privacy & Data Handling for the full retention story.
Where to go next
- §6 Curating the Batch Roll-up · Selection, merges, labels, and reconciliation against your repository
- §7 Delivering & the ArchivesSpace Integration · Readiness checks, ArchivesSpace sends, and the facts that ride along