Source shape
The primary source is a 502-line Markdown document containing research-style narrative, repeated tables and 69 embedded base64 PNG references. The later expanded catalogs contain clean text versions of most image-substituted values. Numbered citation markers do not resolve to a bibliography.
The application never ships the raw Markdown or image payloads to visitors.
Normalization
The curated dataset contains 43 alcohol products, 29 tobacco products, 61 explicit brands, 599 sourced portfolio records, 66 provenance records, 14 leadership records and 12 articles. Names become stable slugs; numeric ranges use separate minimum and maximum fields; display labels remain distinct.
Duplicated rows are reconciled against the expanded master tables. Malformed or contradictory source fields remain traceability notes rather than silent guesses.
Full catalog products and portfolio references are separate. An official global listing can document a product name without proving current state-level availability, pack size, price or technical measurements.
Measurements
ABV is stored as a numerical range where appropriate. Tobacco nicotine and tar values are nominal machine-tested yields, not an individual absorbed dose. The source master table does not provide usable carbon-monoxide values, so the public record says “not provided.”
Terms such as “light,” “mild,” “smooth” or “clean” are retained only when they are part of source-era brand or construction language. They do not establish reduced health risk.
Pricing
All prices are indicative source-era observations. Indian state excise, retailer, date, pack or bottle size, import status and tax policy can produce large variation. USD values are preserved as the source dataset’s comparison equivalents, not live currency conversions.
Market leadership
Leadership records separate sales volume, recognition, historical influence, premium positioning and editorial benchmark status. Because claims can change and source citations are unresolved, they are not presented as live audited rankings.
Storage architecture
The first release uses typed static data behind repository interfaces. That suits a read-heavy editorial dataset, keeps content available during SSR and avoids inventing a D1 production binding. The repository boundary allows a later D1 adapter without rewriting pages.