ToS-100 contains 100 Terms of Service documents from online platforms, splitinto 20,417 clauses. Each clause is annotated for five categories of potentialunfairness: arbitration (A), unilateral change (CH), content removal (CR),limitation of liability (LTD) and unilateral termination (TER). A clause maycarry none of them (18,843 clauses, 92.3%) or several at once, so the task ismulti-label rather than multi-class. This deposit is the corpus of Ruggeri et al. (2022), itself built on the ToScorpus of Lippi et al. (2019), redistributed as three CSV files for Assignment 1of the Natural Language Processing course (Prof. Paolo Torroni, University ofBologna, a.y. 2026-2027). Files: train.csv 80 documents, 15,837 clauses validation.csv 10 documents, 2,548 clauses test.csv 10 documents, 2,032 clauses Columns: document_ID, document, text, A, CH, CR, LTD, TER. The split is by document, not by clause: all clauses of one contract stay inthe same split. Clauses of a single contract repeat each other almost verbatim,so a clause-level split would leak the test set into training. Positive clauses per category (train / validation / test): A 75 / 20 / 11 CH 268 / 43 / 33 CR 165 / 32 / 19 LTD 504 / 75 / 47 TER 318 / 59 / 43 Text is lowercased and tokenized in Penn Treebank style: brackets appear as-lrb- / -rrb-, quotes as `` and '', and clitics are detached (mozilla 's,do n't). The knowledge-base columns of the original release (*_targets), which point tofree-text legal rationales rather than to spans over the clause, are notincluded. Please cite the original works: Lippi et al., 2019. CLAUDETTE: an Automated Detector of Potentially Unfair Clauses in Online Terms of Service. Artificial Intelligence and Law. Ruggeri et al., 2022. Detecting and Explaining Unfairness in Consumer Contracts through Memory Networks. Artificial Intelligence and Law.