Skip to main content

Snorkel AI Data Development Platform v25.11 (STS) release notes

Snorkel is introducing annotation tasks, a major upgrade to the platform’s data labeling workflow. Annotation tasks replace the legacy batch system with a more structured, scalable, and role-aware approach that improves efficiency and transparency across annotation projects.

This release strengthens the foundation of high-quality data development by giving annotators a clearer view of their assignments, providing reviewers with a consistent and intuitive workflow, and enabling administrators to manage and monitor annotation at scale. These enhancements streamline collaboration, reduce operational overhead, and support more reliable training data pipelines for both traditional ML and AI-native workloads.

Annotation tasks are available starting in v25.11. Earlier versions remain available, but batch-based workflows do not migrate forward and require adoption of the new task-based system.

This release also introduces new documentation for annotation tasks, including migration guidance, feature comparisons, and step-by-step how-to articles. We extend our thanks to all contributors, especially Alex Scheller and Jennifer Lei, for their efforts in ensuring the clarity and quality of the documentation.

Breaking changes​

SDK​

  • Removes the in-platform notebook interface to simplify the development experience and reduce maintenance overhead. Use external notebook environments such as Jupyter or VS Code for SDK workflows.

Deprecations​

Annotation Batches​

  • Annotation batches are now legacy. Use annotation tasks for creating, assigning, and managing annotation work.
    For more information, see the Annotation Tasks Overview.

Features and improvements​

Annotation​

Improved annotation task workflow​

  • You can now mark reviewed data points as complete through the SDK, enabling full programmatic control of annotation task workflows.
  • The annotation task data viewers now support consistent filtering, allowing you to quickly find relevant records across question types.
  • The reviewer workflow now uses an intuitive commit flow that works consistently across all annotation question types.
  • Annotations are now scoped to the specific task, so you only see annotations for the task you are working on.

Enhanced data viewing and interaction​

  • You can now view, collapse, and copy JSON fields directly in the annotation viewer, making it easier to review structured or nested data.
  • You can now submit your annotations for review directly from the record view, with clear status indicators and timestamps to help you track progress.

Annotator experience updates​

  • The homepage now highlights all of your assigned annotation tasks, improving navigation and onboarding for new annotators.
  • Annotators now see a simplified sidebar containing only Annotation, reducing distractions and making it easier to focus on active tasks.
  • The Annotation page now shows task details, completion progress, and the number of remaining data points across datasets.

Admin and task setup improvements​

  • Admins can now use Next buttons to return to the task hub after selecting data points or assigning annotators. Live counts indicate how many items were selected or added.
  • Admins can now view real-time status indicators—Unassigned, Assigned, Ready for review, and Completed—in the Annotator Assignment Table.
  • Admins can now select all data points across paginated tables while keeping applied filters, reducing manual multi-page selection effort.
  • Admins can evenly distribute selected data points among annotators with a single action.

Bug fixes​

Annotation​

  • Updates button and font sizing across multiple annotation surfaces for better visual consistency.
  • Fixes an issue where collection cards showed hover behavior even when no action was available.

Known issues​

Annotation​

  • Ground truth updates may not appear correctly for trace datasets in the traces view.

Data management​

  • Some datasets show size discrepancies between the actual file size and what appears in the UI.