A separate release of source scans
Lebanese Court Scans was released on Hugging Face on 30 October 2026. Its dataset card records 44,766 original court-decision PDFs and 274,936 pages. This is a separate publication from the structured legal tables. It gives document researchers access to the scanned material behind court records.
Indexes preserve a route back to the PDF
The documents and pages Parquet configurations load indexes, rather than every PDF. Records retain source links, ruling identifiers, hashes and available fetch dates. The schema explains how to download and verify an individual file. The release is a static snapshot, without a fixed update schedule or a claim to contain every Lebanese judgment.
Unlabelled material needs careful interpretation
The archive mixes handwriting, typewriting, print, stamps and signatures. It has no transcription labels or trained model weights. OCR scores are triage hints. Neither a handwriting-only benchmark nor reviewed reference transcriptions should be inferred from this release.
Read the rights guidance before reuse
The scans are not anonymised and may contain sensitive case details. Their source is the Lebanese University Centre for Research and Studies in Legal Informatics. Retain attribution and source links. The dataset licence is other; MIT applies only to its code. Read RIGHTS.md before choosing a downstream use.