Quran Recitation Event Annotation: A Competition with Private Human Evaluation

A Quran transcript checker must distinguish unresolved mistakes from harmless repetition, successful repairs and accepted spelling differences. We propose a competition that supplies ayah-split transcripts and references, leaves method development to entrants, and scores only their final event predictions against private human annotations. The prepared training release contains 127 unlabelled recording cases and 888 input units, including 44 newly collected cases; a 23-recording review sample comes entirely from training. The private evaluation set contains 100 human-approved cases, 348 units and 162 events across ten labels. An executable label-and-span evaluator and evaluated baselines accompany the task. Agents are permitted, not required, and intermediate iterations are not ranked.