Skip to main content

​
Download

A curated sample of 1,000 recent approved COLA records with associated label-image metadata, OCR text, and extracted barcodes. Download cola-sample-pack-v2.zip No account or API key required.

​
Contents

The ZIP contains three CSV files plus a README and checksum-bound manifest:

​
Relationships

​
Key columns

cola.csv includes 60+ columns. Highlights:
  • TTB_ID — unique identifier for each COLA approval
  • BRAND_NAME, PRODUCT_NAME — the product
  • PRODUCT_TYPE — Wine, Malt Beverage, or Distilled Spirits
  • LLM_CATEGORY, LLM_CATEGORY_PATH — Taxonomy 2.0 category and canonical variable-depth path; a classification can stop at a non-leaf node
  • LLM_PRODUCT_DESCRIPTION — natural language product description from label reading
  • OCR_ABV — alcohol by volume, extracted via OCR
  • BARCODE_VALUE, BARCODE_TYPE — primary barcode from the label
  • MAIN_TTB_IMAGE_ID — identifier for the primary label image. Retrieve a current, short-lived main_image_url from authenticated GET /colas/{ttb_id}; do not persist image URLs.
The archive README and manifest.json identify the exact released taxonomy and classification release, and checksum every included data file. For reclassified records, Taxonomy 2.0 uses label imagery, bounded OCR, and relevant TTB metadata; conservative fallback outcomes remain possible when the evidence or processing result does not support a new placement. Other LLM_ enrichment fields may come from earlier task-specific extraction.

​
Full dataset

The sample represents a small slice of the full COLA Cloud dataset:
  • Millions of COLA records (back to 2005)
  • 5M+ label images
  • 575K+ extracted barcodes
  • Updated daily (~2,500 new approvals per week)
For full access, see the REST API, Snowflake data share, or contact us. See Category Taxonomy for classification semantics and legacy-path transition guidance.