• Window: Czech Rohlík catalogue observed on 11 October 2026, with acquisition timestamps in UTC. Catalogue membership does not establish current availability.
• Grain: one row per retailer product identifier, including food and uncertain classifications. The identifier is not a barcode or GTIN.
• Sources: 19,920 product URLs discovered and 19,917 pages acquired. TheMealDB supplied 790 recipes and 8,152 nonempty ingredient positions.
• Matches: 388 ingredient positions have candidates across 225 products. Multiple products can match one ingredient position.
• Quality: five declared checks passed without source truncation. The yellow cayenne chilli, Lipánek vanilla quark and Klorane dry shampoo pages were not acquired successfully. 13,259 products require review.
• Cadence: weekly updates are configured. The registered Toprecepty expansion still needs its first full build and refresh verification.

• product_id, product_name: retailer identifier and original product heading.
• standardized_name_cs, search_text: inferred standardized identity and retrieval text retaining the original title.
• ingredient_key, food_scope, cut, form, preparation_state, storage_state: inferred identity, scope and attributes. Exposed inferred attributes require confidence of at least 0.85.
• classification_confidence: lowest identity and scope confidence, on a 0–1 scale. Confidence is not calibrated accuracy.
• classification_json, classification_model, taxonomy_version, classification_input_hash, review_status: raw decisions, served model, vocabulary version, input hash and review flag.
• pack_quantity_text, description_text, ingredients_text, allergens_text, nutrition_text: extracted publisher text. Nutrition retains the original units and reporting basis.
• page_text: captured page text for extraction audit, including unrelated recommendations.
• source_url, fetched_at, snapshot_date: captured product URL, UTC acquisition timestamp and UTC date.
• image_url, image_urls: primary image URL and ordered gallery URLs.
• recipe_matches: ingredient candidates retaining original names, measures, constraints and evidence. Quantities and servings are not normalized.
• matched_recipe_count, matched_ingredient_count: counts of distinct matched recipes and matched ingredient positions per product.
• recipe_match_coverage, recipe_matching_integrity_verified: observed recipe coverage and classifier-answer integrity indicator. Coverage counts do not prove acquisition completeness.

Missing extracted text and images mean unknown, not zero or absence. Inferred labels retain unknown or not_applicable where appropriate, and allergen text preserves reported traces without inferring allergy safety.

• Searching Czech grocery products by original names and inferred attributes.
• Inspecting ingredients, allergens and original nutrition text.
• Exploring candidate product matches for international recipe ingredients.
• Auditing classifier decisions and products requiring review.

Rohlík publishes the Czech product catalogue and TheMealDB supplies the international recipes. No open reuse licence is asserted for Rohlík or the combined result, and recipe candidates do not establish availability, allergy suitability or substitution equivalence.