2026-02-02 · Rika Sato
Labeling queues that respect annotator wrists
human-loop · operations
Most labeling backlogs fail because throughput metrics ignore ergonomics. We map keyboard paths the way product designers map checkout flows—fewer modal hops, predictable shortcuts, and batch sizes that match how long someone can focus before a break.
Cohort groups redesign a queue using real (sanitized) screenshots. Some teams add a “reason skipped” field; others collapse adjudication into a single pane. Diversity of approach is the point.
We document outcomes with modest claims: time-to-label dropped modestly, wrist complaints fell, reviewers stayed in the tool longer. No miracle percentages—just observable comfort.