About the Taxonomy

The Culinary Atlas uses a numeric coding schema to classify dishes along multiple axes: primary cooking method, macro ingredient category, preparation tradition, and more.

Each dish receives a taxonomy_codes JSON object (stored on the dishes table) with keys like main_method, macro, and prep. These codes also determine the quilt color index used for visual display in the Atlas.

The taxonomy was created for the live documentary German Soul, to research and discover connections between Soul Food, and food from Germany, West Africa, and the Caribbean. Once fully developed for this use case, the taxonomy will join the atlas in becoming a powerful resource for food scholars worldwide.

Data Collection

The Southern Culinary Atlas collects data through three primary channels:

• Editorial seeding — the project team contributes canonical dish records and editorial recipes as a baseline corpus, particularly for dishes with thin oral documentation.
• Oral History Quiz — a structured eight-prompt quiz in the Interactive Suite elicits personal food memories. Responses are anchored to a self-reported county and linked to dishes where identified.
• Open contribution — contributors may submit recipes and oral histories outside the quiz structure via a free-entry form. These follow the same moderation process.
• Dish Scout (Dish Flash) — a gamified recognition interface generates binary dish-recognition responses (yes/no) per county, producing validation signal for existing records.

Moderation Process

Every contributed record passes through a moderation queue before publication. Moderators are community members who applied and were approved by the editorial team. The moderation process reviews:

• Content accuracy — obvious errors or anachronisms in provenance claims
• Geographic plausibility — contributions are anchored to a claimed county; implausible combinations are flagged
• Cultural sensitivity — content that misrepresents or appropriates oral tradition is declined with explanation
• PII detection — moderators are trained to catch and remove identifying information before approval

Records declined by moderators are not published and not included in exports. The moderator who approved a record and the timestamp are retained internally but are not exported.

Pseudonymization

Contributor PII is held in a separate, access-controlled database table (contributors) that is never exposed through the public-facing API or export endpoints. Every contributor is assigned a pseudo_id (a stable, non-reversible hash) which appears in exported recipe and oral history records in place of any identifying field.

The mapping between pseudo_id and real identity is held only in the admin-accessible table. No export format, endpoint, or public interface exposes this mapping. Geographic granularity in exports is county-level only — no street addresses, zip codes, or sub-county identifiers are published.

License Terms for Downloaders

Each recipe record carries a license assigned by the original contributor at time of submission. The available licenses are:

• CC0 — public domain dedication; no attribution required
• CC BY — attribution required; derivative works permitted
• CC BY-NC — attribution required; non-commercial use only
• All Rights Reserved — the contributor retains copyright; contact the editorial team for use

Oral histories default to CC BY-NC unless the contributor specifies otherwise. Dish editorial records are CC BY (the platform holds copyright). The license column is present in all recipe exports. Downloaders are responsible for respecting per-record licensing.

Limitations and Biases

Researchers using this dataset should be aware of the following:

• Geographic skew — early data collection is concentrated in Eastern North Carolina, the Lowcountry of South Carolina, and Atlanta metro. Gulf Coast and Appalachian coverage is thin.
• Survival bias in oral history — the quiz format privileges memories that are already narratively formed. Everyday, unremarkable foodways are underrepresented.
• Language — the platform is English-only in V1. Spanish-language and other-language contributions are not yet supported.
• Network recruitment — early contributors were recruited through the documentary production's existing community contacts, introducing a network homophily effect.
• Taxonomy is editorial — dish classification is performed by the editorial team using Justin's schema. It is not crowd-sourced and reflects editorial judgment, not community consensus.
• Sandbox data — the dataset in this release is editorial seed data. Statistically meaningful analysis should await a full contributor dataset.