CLEAR Corpus (CommonLit Ease of Readability)

The CommonLit Ease of Readability (CLEAR) Corpus is a collection of around 5,000 reading passage excerpts levelled for grade 3 to grade 12 readers, built by CommonLit, a non-profit education technology organisation, with Georgia State University, and maintained by Scott Crossley. Each excerpt draws from over 250 years of writing across two genres (literary and informational) and carries a rich set of labels: multiple readability indices (Flesch Reading Ease, Flesch-Kincaid Grade Level, Automated Readability Index, SMOG, New Dale-Chall, and the corpus-specific CAREC, CAREC_M, CARES and CML2RI), plus publication year, genre and other metadata. Its defining feature is a human criterion: teacher ratings of how difficult each text is for student readers, which give a measured difficulty score rather than a formula alone.

You access it as a single spreadsheet (CLEAR_corpus_final.xlsx) in a small public GitHub repository, with a mirror on Kaggle, so ingestion is a one-file download and a parse rather than a crawl or an API.

For RAG the corpus is small but unusually well labelled, which makes it useful as an evaluation and calibration set rather than a knowledge base: you can test whether a retriever or a generation step holds to a target reading level. For an AI tutor the value is direct: the excerpts let you match text difficulty to a learner's reading level, sequence practice reading by grade, and train or validate a model that predicts readability, so passages and explanations are pitched correctly and progression can be checked against a measured scale.

The licence is the crucial caveat. The dataset is released under CC BY-NC-SA 4.0: non-commercial, share-alike and attribution required. That rules out any paid or commercial tutor without separate permission, so treat it as a research and non-commercial resource. It complements the teaching content we list (openstax, ck-12, khan-academy, siyavula) and the standards spines (common-core, ngss): those supply graded lessons and competencies, while CLEAR supplies a measured reading-difficulty signal to level them by.