What this atlas is, and is not

Scope

Thought Atlas maps recorded human culture: surviving written works and the material world around them, from among the earliest administrative tablets at Uruk, c. 3400 BC, through late antiquity, c. AD 500. Every entry rests on a text, an inscription, or an object with writing on it.

Symbolic and material culture before writing, cave art, portable art, musical instruments, burials, counting tokens, Göbekli Tepe, is out of scope by design, not by oversight. It is older, richer, and in places better evidenced than anything on this site, and it deserves treatment on its own terms rather than a token gesture inside a project built around the written record. The atlas does not claim that culture, or recoverable human thought, began with writing. It claims only to map what writing preserved.

Chronological conventions

Dates use BC and AD, not BCE and CE. There is no year zero: 1 BC is followed directly by AD 1. Centuries are numbered on the same convention, so the 34th century BC runs from 3400 to 3301 BC and the 1st century AD runs from AD 1 to AD 100. A date is assigned to a century by rounding toward the more distant end of its span from year zero, the same rule in both directions across the BC/AD boundary.

Where a work's date is disputed or spans a range, the entry's displayed date states that range or approximation directly (for example, "c. 9th-5th century BC") rather than collapsing it to a single year for tidiness.

Four dates, one field

A work can carry as many as four distinct dates: when its events are set, when it was first composed, when it reached the form that survives, and when the earliest surviving copy or inscription was made. Conflating any of these with another is a common source of false precision, and this site does not always avoid it: the corpus's date field represents composition, qualified by a precision value (exact, decade, century, or range) that states how tightly that composition date is actually known.

Where the earliest surviving witness, manuscript, inscription, or fragment, is known and materially later than composition, it appears as its own line on the entry (attestation.witness), so a reader is never left assuming a medieval codex proves a Bronze Age date. Historical setting and final redaction, where they diverge from composition, are stated in the entry's prose rather than in a separate structured field; the corpus does not yet carry dedicated fields for every work with all four dates distinct, and this is logged as a known gap rather than papered over.

Certainty labels

Where shown, an entry's status chips read one of five labels, applied separately to composition and to authorship:

• Secure - the date or attribution is not seriously disputed among specialists.
• Probable - the balance of evidence favors this reading, without full consensus.
• Contested - specialists are genuinely divided, with no settled majority view.
• Traditional - the attribution is carried by tradition rather than established by evidence internal or external to the work.
• Range-uncertain - the date range itself is too wide, or too disputed, to state as a single secure reading.

These labels are additive and applied progressively: the brief's language-risk audit governs every entry's prose regardless of whether it yet carries a chip, and the absence of a chip on an entry is not a claim of certainty. It means the label has not been added yet.

Inclusion criteria and proportionality

An entry requires a surviving text, inscription, or archive, or a securely attested script even where its language remains undeciphered. Lost works known only by reputation or citation are not entered on their own; where they bear materially on a surviving work's context, they are named in that work's prose instead. Undeciphered scripts are included and are not described as unreadable where their numerical systems, document structure, or recurring signs are partly understood.

Archives are treated as a single entry rather than one per tablet: the Ebla Palace G archive and the Amarna letters are each one entry describing a body of material, not a few thousand. Well-represented authors and traditions do not automatically receive a new entry for every additional surviving work; a major archive, script, or textual tradition still absent from the corpus generally takes precedence over a third or fourth entry for an author already covered.

Source hierarchy

Sources for text, dating, and links are preferred in this order:

1. A digital corpus or critical edition (CDLI, ORACC, ETCSL, CText, Sefaria, Perseus/Scaife).
2. A museum object record (British Museum, Met, Penn Museum, and comparable institutional collections).
3. An excavation publication or specialist university project.
4. A peer-reviewed scholarly overview.
5. A general encyclopedia, used only where no better source exists, and never presented as a "read the text" link.

Wikipedia and general encyclopedias are acceptable for background but not offered as full-text links. Shadow-library mirrors are excluded outright regardless of scholarly quality; a published site does not link infringing scans. A lending-gated archive.org item is used only as a last resort, when no open institutional text exists, and is labeled honestly rather than presented as freely readable.

Translation and naming

Titles are given in a conventional English form, with the original-language title shown alongside where the entry's writing system renders reliably. Place names on the atlas are short toponyms, at most one parenthetical alternative, not a caveat sentence; uncertainty about a composition site belongs in the entry's prose, stated once, not repeated in the map label. Regional and period names follow the terminology in current use among specialists in that field rather than an older or more familiar label where the two have diverged.

Revision policy and corrections

This atlas is a working document, not a finished one, and it will contain errors that have not yet been caught. Corrections, disputes, and requests for a specialist's second look go to the Board, the site's public commentary surface; a claim flagged there as needing verification is tracked rather than silently left in place. Where a change is made in response to a correction, the site's editorial changelog records what changed and why.

Historical basemaps

Political borders on the atlas and century pages are approximate scholarly reconstructions from historical-basemaps (GPL-3), reprojected and simplified for this site. They are drawn to the nearest available anchor year to a given century or scene, not to that century's own boundaries precisely, and are labeled as approximate throughout.

Reconstruction plates

Some galleries and city cards carry photorealistic images labeled "AI reconstruction". These are modern images generated with an image model (gpt-image-1), each grounded in a named piece of evidence: an artifact-based plate is generated from the photograph of the specific museum object it names, and a city plate from the published excavation record of the site at a stated period. A plate with no such grounding is not made.

Faces follow a stricter rule. A person is reconstructed only where a contemporary or near-contemporary likeness survives, a portrait sculpture, relief, coin portrait, or painted portrait of that person. No face is invented for anyone whose appearance the record does not preserve; idealized or generic sculpture can ground an unnamed "person of the period", never a named individual.

Every plate is labeled. The tag "AI reconstruction" appears directly beneath the image, and a provenance line beneath it names the artifact or excavation evidence the image rests on, so a reconstruction is never mistakable for a photograph of an object or a place. These images are interpretations, not records: they show one defensible reading of the evidence, rendered by a model, and they carry no evidentiary weight of their own. Where the plate and the object disagree, believe the object.