Hadith Attribution Corpus: An Expert-Annotated Arabic Dataset for Attribution-Level Classification (R/RS/S/SH)

An expert-annotated Arabic corpus of 2,077 hadith narrations labelled by attribution level — whether the content of a narration belongs to the Prophet, to a Companion (Ṣaḥābī), or to both. Narrations are drawn from al-Bukhārī (al-Ṣaḥīḥ and al-Adab al-Mufrad), Muslim, Abū Dāwūd, al-Tirmidhī and al-Nasāʾī, and labelled into four categories: R (Prophetic direct speech), RS (mixed Prophetic and non-Prophetic discourse), S (Companion/non-Prophetic discourse), and SH (Companion- or narrator-formulated Prophetic practice or ruling). All narrations were annotated by a single hadith specialist following the written guideline included in this deposit. Reliability was quantified by an intra-annotator test–retest study on a stratified sample of 312 narrations: overall Cohen's κ = 0.81 (raw agreement 86.5%). The dataset accompanies a journal article currently under review.