Quality methodology
Inter-annotator agreement
Every dataset entry is labelled by ≥3 independent annotators. Fleiss’ κ and pairwise span F1 are published per release.
Low-agreement cases
No merged entries yetAwaiting first releaseOnce merged dataset entries ship, low-agreement samples appear here for inspection — each annotated with per-annotator votes and the final resolution.
Agreement over time
Weekly average κ and the volume of entries merged each week.
No merged entries yetAwaiting first releaseNot enough history yet — a trend needs at least two weeks of merged entries. Come back after the next release.