Skip to main content

April 2025

Continues from the 17 April 2025 release. Projects can now manage their own pipeline steps, and a role permission now decides who may clear errors on a task.

Week of 21 April​

Prediction & processing​

  • Auto-process on warnings: a project option to let tasks continue automatically when validation produces warnings but no errors.
  • Stored classification results: the classification shown on a task is stored, so it is the one used for the rest of processing.
  • File type from content type: incoming files are handled according to their declared content type, so mislabelled uploads behave more predictably.
  • Repeat anonymisation: documents can be anonymised again, and the same name always gets the same placeholder.

Administration & access​

  • Permission to clear errors: clearing an error on a task now needs a role permission.

Configuration​

  • Label descriptions: project and entity labels can have a description, shown to reviewers and used as an instruction for the AI model.

Week of 28 April​

Prediction & processing​

  • Pipeline steps: add, reorder and configure pipeline steps in the interface or through the API.
  • Prediction settings: prediction behaviour now has its own page in project settings.

Export​

  • Re-export a single task: send one task to the export connector again without re-exporting its whole upload. Failed exports have a retry action.

Interface​

  • Blocked tasks in table view: the table view now marks tasks that are blocked from processing.
  • Faster upload lists: the upload list refreshes more often in busy folders and less in idle ones.

Fixes​

  • Anonymising large documents no longer fails.
  • Manual uploads now land in the right folder.
  • Project counts, "classify with label" and email delivery settings save correctly again.
  • Task metadata and document images load without errors again.
  • Creating an organisation no longer fails.