Clouda OCR Benchmark v0.1.0

Clouda OCR Benchmark v0.1.0 is the first public release of the project's reproducible distorted Arabic OCR benchmark.

Highlights:

• 177 canonical distorted Arabic benchmark samples
• 6 models with complete 177/177 valid coverage
• HunyuanOCR-1.5 ranks first by normalized Arabic CER
• Additional dots.mocr run: 177 statuses, 169 valid outputs, 8 failures
• PaddleOCR-VL-1.6 preserved as failed/unranked evidence
• Common 169-sample successful subset included
• Reproducibility metadata, methodology, checksums, and public audit artifacts included
• Dataset redistribution reviewed conservatively; assets with uncertain redistribution rights are withheld

Important:
Results are specific to this benchmark and do not imply universal OCR performance.