Summary
This guide explains how to configure task assignment for efficient, reliable labeling in Label Studio Enterprise. It covers when to use Auto vs. Manual distribution, setting overlap and low‑agreement escalation, skip queue and task reservation behavior, task sampling, review routing, and sharing task links for team calibration. You’ll also see scenario tips for choosing Manual vs. Auto, plus where to add members and publish. All steps map directly to Enterprise settings pages so you can apply them quickly.
Optimal Workflow steps
- Choose a task distribution mode
- Auto distribution (recommended): Annotators can start immediately after publish; the system enforces overlap and manages the queue. Supports task reservation and skip behavior. See Auto distribution and Project settings.
- Manual assignments (strict routing): Annotators only see assigned tasks; assign via Data Manager for per-team subsets. See Assign annotators to tasks.
- Configure overlap and low‑agreement escalation (quality loop)
- Overlap: In Auto distribution, set “Annotations per task minimum” to the number of independent labelers required per task. See Project settings – Quality.
- Low‑agreement strategy: Choose the agreement metric, threshold, and maximum additional annotators to auto‑escalate low‑agreement tasks. See Project settings – Quality.
- Set skip queue behavior and task reservation (lock timeout)
- Skip handling: Choose whether skipped tasks requeue to the same user, requeue to others, or are ignored until reassignment.
- Task reservation: Set how long a task stays reserved for an annotator before it returns to the pool.
- See Auto distribution (skip + reservation) and Project settings.
- Pick a task sampling strategy
- Select sequential, uniform, or uncertainty sampling to control what gets labeled next (useful for active learning). See Enterprise features.
- Configure review routing
- Decide when tasks appear in review (e.g., after required annotations).
- Optionally require manual reviewer assignment. See Project settings – Review.
- Add members and publish the project
- Add annotators/reviewers at the project or workspace level, then publish so annotators can start. See Set up a project.
- Decide which scenarios use Manual vs. Auto
- One‑pass labeling: Auto + overlap = 1.
- Multi‑annotator overlap: Auto + overlap = N; optionally enable low‑agreement escalation.
- Strict per‑team routing: Manual assignments via Data Manager; each team only sees its tasks. See Assign annotators.
- Calibrations and deep links to specific tasks
- Share a specific task from the Data Manager using a link with both task and tab parameters (only assigned users can open in Manual; Auto requires the user be eligible/assigned). See Deep linking and Project settings.
- Monitor, adjust, and scale
- Use dashboards and agreement metrics to monitor throughput and quality; tune overlap, skip behavior, sampling, and escalation thresholds. See Quality workflows and Annotator dashboard.
Tip: If each task must be labeled by specific people (e.g., A and B), use Manual and assign both. If “any two” annotators are fine, use Auto with overlap = 2.
More resources
- Project settings (Enterprise): Annotation, Review, Quality
https://docs.humansignal.com/guide/project_settings_lse - Auto distribution, reservation, skip behavior, sampling
https://labelstud.io/guide/enterprise_features.html#auto-distribution - Create and configure projects (add members, publish)
https://docs.humansignal.com/guide/setup_project.html - Assign annotators to tasks (Data Manager)
https://docs.humansignal.com/guide/manage_data.html#Assign-annotators-to-tasks - Quality and agreement workflows
https://docs.humansignal.com/guide/quality.html - Deep linking (share precise task/annotation locations)
https://labelstud.io/guide/enterprise_features.html#deep-linking