Skip to main content

Reviewing AI Classifications

After processing, you review what the AI extracted. The Archiver is explicit that you're in charge — the model proposes, you dispose.


Where to find it

From the accession page, the Items tab is the review surface.

Items tab


The toolbar

ControlWhat it does
List / table view toggleCards vs. spreadsheet-style
Select allTick every visible row
All / FlaggedShow only items the AI flagged for review
Expand / CollapseOpen / close every row at once
Accept by categoryBulk-approve every item of a given category
All fields dropdownFilter which fields show when rows are expanded

The REVIEW summary strip beneath shows progress per category (e.g. Artefact 0/1, Audio 0/2, Video 0/1) — that's how many in each category you've accepted.


What you're reviewing per item

When you expand a row, you see every field in that item's metadata schema. The schema comes from Settings → Data Model — different categories have different fields.

Item review

For an artefact, that's typically:

  • date — date or date range
  • title — short descriptor
  • condition — narrative conservation statement
  • materials — controlled list (resin, paint, paper, etc.)
  • dimensions — measurements
  • description — full description
  • object type — controlled vocab (figurine, badge, programme, …)
  • inscriptions — any text on the object
  • pii detected — whether the item contains personal data
  • access rights — Public / Restricted / Closed
  • creator maker — creator or maker of the object
  • date certainty — how confident the model is in the date
  • content warnings — sensitive-content flags
  • provenance notes — how the object came to the repository

Documents, photographs, audio, and video have their own schemas — see Settings → Data Model for what's active in your organisation, and Data Model to customise.


Editing fields

Click any value to edit it inline. Pills (authority-linked fields like subjects, people mentioned) open a small picker:

  • Type to add a value — the picker offers matches from the bound vocabulary (LCSH, FAST, Getty AAT, TGN, etc.)
  • Click an existing pill to remove or swap the linked authority

The AI's first proposal is editable; you're never locked in.


Accepting items

Once a row is reviewed, click Accept at the end of that row. The row's status flips to accepted and counts toward the per-category progress.

Accept by category lets you bulk-approve everything in one category in a single click — useful when the AI did well on, say, all the photographs but you want to look at each video individually.


Flagging for later

If you're not sure and you want to come back, leave the row unaccepted. The Flagged (n) filter surfaces everything still outstanding, so you can return and finish without sifting.


After review

When all items in the accession are accepted, the accession's status flips to Complete. You can:

  • Run Collection Analysis (Analysis tab) for a top-down summary of the batch
  • Review linked authorities (Authorities tab) — verify the extracted subjects, people, and places resolved correctly
  • Export via the Export panel at the bottom of the accession page

Tips

  • Trust the model on routine fields, scrutinise the headline ones. Title, description, and access rights deserve a careful read. Date certainty and PII flags are usually right.
  • Subjects in particular are worth checking. A wrong subject heading propagates into your finding aid and into Search.
  • Don't accept then change your mind. You can re-open and edit accepted items at any time — Accept is a workflow signal, not a lock.