Human Food Atlas

Human Food Atlas began with a simple question: what would the history of humanity look like if food were the organizing principle? Food connects biology, environment, technology, migration, trade, culture and everyday life more intimately than almost any other subject. Yet its history is usually scattered across archaeological reports, historical texts, nutritional databases and separate accounts of individual regions and periods. HFA brings those fragments onto a single spatiotemporal model of human food history, allowing foods, dishes, technologies and populations to be explored together across geography and time.

The project combines editorial research with structured data and AI-assisted historical reconstructions grounded in current archaeological, historical and scientific literature. Visual scenes are treated as evidence-based reconstructions rather than illustrations unconstrained by the record: documented facts are separated from scholarly inference and reasonable reconstruction, and sources accompany substantive historical claims. The result is intended to be both exploratory and rigorous—a continuously evolving atlas in which a meal, ingredient or cooking technique can be followed through thousands of years of human history.

Method

Each entry in the atlas is built the same way. A moment is fixed in place and time — a settlement, a port, a kitchen, a famine — and everything the record supports is assembled around it: the foods available there, the techniques used to produce and prepare them, the trade that carried them and, where the evidence allows it, an estimate of the nutrition on the table. The written account, the structured data and the reconstructed scene are three views of one underlying record, kept consistent with one another rather than produced independently.

Evidence and reconstruction

Reconstruction here is a discipline, not a licence. A documented fact — a carbonised loaf, a cuneiform ration list, an isotope signature — is stated as one. Scholarly inference is presented as inference. Where a scene requires details the record cannot supply, they are chosen to be consistent with everything the record does supply, and nothing in a scene may contradict the evidence. Verified sources are published with an entry as the sourcing pass reaches it, and entries state their limitations — what is known, what is inferred and what is honestly uncertain. Entries are revised as the literature moves.

Scope

The atlas runs from the first evidence of hominin food processing, more than three million years ago, to the present, and follows food wherever people carried it. It is a living dataset: the figures below are a reading from it, not a description of it.

Table 1. Current holdings.

Measure	Count
Entries	6,787
Reconstructed scenes	4,352
Collections	6
Historical epochs	16
Countries represented	195
Population centres mapped	903

How the atlas is made

Human Food Atlas is a small independent project, and it is made the way it describes itself: as a collaboration between editorial judgment and machine labour. A single editor sets the questions, the standards and the taste. AI systems work under that direction — researching, drafting reconstructions, generating scenes, estimating nutrition — and their output is constrained by structured rules and passed through automated checks and editorial review before it is published. The division of labour is deliberate: machines are good at holding thousands of records to one standard; an editor is what keeps a standard worth holding.

Corrections

An atlas of this scope will contain mistakes, and the project would rather hear about them than defend them. Every card carries a control for reporting an issue — a wrong date, a doubtful ingredient, an anachronism in a scene. Reports are read, and a report that holds up will change the record.