Help - ProofCheck

Getting Started

ProofCheck helps you verify alcohol beverage labels against TTB (Alcohol and Tobacco Tax and Trade Bureau) regulations. There are two workflows:

Single Label Verification — Upload one label image, extract text, and compare against expected data.
Batch Verification — Upload multiple label images and process them all at once.

Single Label Verification

Step 1: Upload a Label Image

Upload a photo of the alcohol beverage label you want to verify.

Supported formats: JPG, PNG, WebP
Maximum file size: 5 MB
For best results, use a clear, well-lit photo taken straight-on

Step 2: AI Extracts Text

The AI model reads text from the label image, extracting fields like brand name, alcohol content, net contents, producer information, and the government warning.

The government warning is automatically checked against the standard TTB text required under 27 CFR § 16.

Step 3: Compare with Expected Data (Optional)

Enter the expected label data (e.g. from your COLA application) to compare against what was extracted. The system normalizes values automatically — for example, "80 Proof" and "40% ABV" are recognized as equivalent.

Batch Verification

Uploading Multiple Images

Drag and drop or select multiple label images at once. All images are processed concurrently for faster results.

CSV Import

Upload a CSV file to auto-populate expected data for each label. The CSV should include:

A fileName column matching your uploaded image file names
Columns for label fields: brandName, alcoholContent, netContents, etc.

Template Feature

Use the template form to fill in data common to all labels (e.g. producer name and address), then click "Apply to All Files" to copy it to every label's form. Individual labels can still be overridden.

Understanding Results

Overall Status

Approved — All fields match the expected label data.
Needs Review — Some fields have minor differences that may still be acceptable. Manual review is recommended.
Rejected — One or more fields do not match the expected data.

Field Statuses

Match — The extracted value matches the expected value.
Warning — Minor difference detected (e.g. case variation). May still be acceptable.
Mismatch — The extracted value does not match the expected value.
Not Found — This field was not detected on the label.

Government Warning Compliance

Every label is automatically checked against the standard TTB government warning text required under 27 CFR § 16. This check runs regardless of whether you provide expected data.

Confidence & Image Quality

Confidence indicates how certain the AI model is in its extraction. A higher percentage means more reliable results.

Image quality is the AI's assessment of the uploaded image clarity. Poor image quality may lead to less accurate extractions.

Tips

Best Practices for Label Photos

Use good, even lighting — avoid shadows across the text
Take the photo straight-on to minimize distortion
Avoid glare, especially on glossy or metallic labels
Make sure all text is in focus and fully visible in the frame

Model Selection

Sonnet — More accurate but slower. Best for final verification and difficult-to-read labels.
Haiku — Faster but may miss some details. Good for quick checks and batch processing.

Supported Input Formats

The system normalizes common formats automatically:

Alcohol content: 40% ABV, 80 Proof, 40% Alc./Vol., or just "40"
Net contents: Compared as text with spacing normalization (e.g., "750ml" matches "750 mL")
Addresses: Abbreviations like St, Ave, Blvd are expanded for comparison