Yutuji: a hand-drawn historical atlas with graded sources

Summary

Yutuji (舆图集, “a collection of maps of the land”) is a Chinese-language series of hand-drawn historical maps. Each issue takes one text or one theme, a monk's itinerary, a treatise on crafts, a city's walls, and maps the places, people and dates the sources name. As of 26 October 2026, 19 of 92 planned issues are published. Behind the issues sits one shared set of records: 1,059 places, 277 people with 646 dated itinerary entries, 580 events, 128 objects and a catalogue of 334 cited works. Every date and fact carries a source string and a grade from A to D. This note describes how the records are made, how the grades are assigned, what the maps do and do not claim, and where the atlas is known to be weak.

1. Scope

The issues are arranged in seven volumes, each defined by a kind of question rather than by period or region: cities and their walls (城郭志), journeys (行役志), goods and trade (货殖志), translation of texts and ideas (重译志), maps and how an age drew its world (舆地志), rivers and canals (河渠志), and passes and roads that the terrain forces travellers onto (关隘志). Periods run from the Neolithic to contemporary craft practice. The published issues are listed, with English abstracts, on the English page.

An issue is a single web page: a series of short chapters, each with a small map, and a full-screen map tool with several modes (places, time, people, and whatever the subject needs). Every issue also feeds the shared records, so that a place or a person mentioned in several issues is one record with one address.

2. Texts and transcription

Quotations are taken from digital texts and checked character by character against them when an issue is compiled. Most classical texts come from the Daizhige corpus (殆知阁古籍库); Buddhist texts from the CBETA edition of the Taishō canon (for Xuanzang, T2053 and T2087; for Faxian, T2085); a few from Wikisource and from library scans. Where a digital base text has an unusual or erroneous character, the quotation keeps it and a note gives the usual reading, so that the quotation can still be found in the base text. Where two texts disagree (the Hanshu against the Weilüe on the roads of the Western Regions, the annals of the Hanshu against its own treatise on the dates of the four Hexi commanderies) both are given and neither is chosen.

Every classical quotation is accompanied by a modern Chinese version (白话). This is a translation, not a commentary: it follows the order of the original, keeps names of people, places, offices, books and reign periods as they are, keeps numbers and measures unconverted, and adds nothing the original does not say. Where a sentence is unclear, it is translated literally rather than explained. Three kinds of text get no modern version: lists of names, modern prose, and sentences that are already plain. There are no English translations of the sources.

Several texts that matter are not available to the editors in a checkable form. Where a claim rests on a modern author's quotation of a source that could not be seen, the citation says so and the grade drops accordingly. The Chinese first uses of “丝绸之路”, for example, are known only through Liu Jinbao's 2018 article, and are graded C.

3. Grades

Every citation has a grade:

The grade describes the relation between a statement and its source, not the truth of the statement. An A means the statement is found in a text written at the time; in the Tiangong kaiwu issue every entry is A because every entry is a sentence of the book, including Song Yingxing's reports of things he had only heard. What counts as “contemporary” depends on the subject, and each issue's notes give examples: for Xuanzang, the Tang biography and the Xiyu ji; for the West Lake houses, which rest mostly on local history and the press, nothing is graded above C, and oral accounts without documentary support are D.

Across the records there are 3,268 citations graded A, 1,157 B, 2,434 C and 182 D. A date calculated from another date (a birth year from an age at death) is B even when the source is contemporary. An identification of an ancient place with a modern one is C unless the identification itself is attested. Legends are kept, graded D and cited, because what later ages believed is part of the history of a place; the walnuts and pomegranates that Zhang Qian is supposed to have brought back are an example.

4. Source strings and the catalogue of works

A source is written as in Chinese scholarship: the title in 《》, then juan and chapter (《宋史》卷四百一《辛弃疾传》). Several sources in one citation are separated by “；”, and a segment that begins with 卷 (“juan”) continues the book of the previous segment. Sources that are not books carry a bracketed mark: 〔今人〕 modern scholarship, 〔方志〕 local gazetteer, 〔寺志〕 monastery gazetteer, 〔碑刻〕 inscription, 〔图录〕 catalogue, 〔数据库〕 database, 〔口碑〕 oral account, 〔本期〕 the editors' own inference (a coordinate estimated from a street name, a year interpolated between two known stages).

A program reads every source string, resolves it to a work in a catalogue of 334 titles with their variant names, authors and dates, and builds one page per cited work listing every passage the atlas takes from it. 305 works are cited on the pages; about 123 source segments are neither a titled book nor a marked source, and the aim is for that number to fall with each issue.

5. Places

A place record has a historical name, a modern name, coordinates, a precision, the place it lies within, and the issues that mention it. Of 1,059 places, 182 are located to a point; the rest are located to the level of a town, a county, a prefecture or a region, or approximately, and 201 are not located at all. Coordinates are modern references; the original datum and uncertainty are retained where they are known. The maps combine ink-line symbols with geographic data: coastlines and rivers are simplified from Natural Earth, with finer modern reference data where an issue needs them. These layers do not reconstruct ancient shorelines or locate individual kilns. Ricci's world is redrawn in the map's own oval projection.

Two rules govern what goes on a map. First, a place the sources name but do not locate is not placed: the Tiangong kaiwu ties its products to 373 place names in 520 sentences, and the names that are only provinces, regions or directions (“the south”, “Min and Guang”), or that the Ming administrative geography does not contain, are listed with the reason and left off the map. Second, links are not guessed. An event is attached to a place only when the name in the event matches a place record exactly, or matches once a modern gloss in brackets is removed; at present 321 of 580 events are linked, and the rest keep their place name as text.

A place inside a city (a house on West Lake, a gate of Hangzhou) has no small map of its own on its page, since the base maps are drawn at the scale of prefectures and would show only the city; it links to the city instead.

6. People and itineraries

A person record has dates, a short statement of who the person was, a short biography, and an itinerary: a list of dated stages, each with a place, what happened there and its own citation. There are 646 such stages for 277 people. Dates are written as the source gives them, with a numeric year for sorting; estimated years are marked 约 (“about”), and where the sources disagree (Xuanzang's year of departure, 629 or 627) both are given and the disagreement goes to the doubts page.

7. Doubts

What the atlas cannot confirm is collected on one page, 存疑, generated from the data. It has three parts: open questions and conflicting accounts (276), places named in the sources but not located (279), and statements graded D (182). Statements with the same reason are shown together, so that fourteen wards of Southern Song Hangzhou whose modern locations are estimates take one line.

8. Objects

128 objects that appear in the issues (silk, lacquer, paper, a folding stool, a canal boat) have their own pages, and 136 of them are drawn. The drawings follow the conventions of archaeological line drawing: ink line in two weights, and exactly one touch of vermilion on each object, placed on the part the text names or, where the text names none, on the part that matters most to its structure. Where the text gives dimensions, the drawing follows them: the Tiangong kaiwu gives the mast of a grain barge as eight zhang and the hull as seven zhang one chi overall, so the mast is drawn longer than the boat. Where a text records how a thing was made or used but not how it looked, it is not drawn.

9. What the atlas does not do

Several things are left out on purpose. Year pages give the Western year and its sexagenary designation but not a reign title, because in periods with several courts the choice of one court's reign title is itself a judgement. Craft processes in the Tiangong kaiwu are not mapped, only the link between a thing and a place. Modern geographic data are used as reference where stated; they do not establish historical routes. Satellite imagery is not part of the published basemaps. Places are not moved to fit the drawing: where an endpoint falls outside a map, it is marked at the edge. Tempting claims that could not be traced to a source are left out: the phrase 东南佛国 (“Buddhist land of the southeast”) is not found in any of the Song and Ming works on Hangzhou that were searched, so it is not presented as a Song saying.

10. Records, ids and versions

The shared records are five JSON files (places, people, events, works, things); each issue adds its own files for what only it needs (the stages of the Grand Canal, the editions of the Haiguo tuzhi). Every record has a permanent id, which is also the address of its page (/place/id/, /person/id/). An id is never reused; when a record is removed or merged its id is kept in a list of retired ids, and its old address redirects. The shared records are exported as JSON and CSV on the open data page, mirrored in a public repository, and archived on Zenodo with a DOI for each release. Changes to the site are logged, with dates, in a journal (刊记) that is also an Atom feed.

11. How the atlas is made

The atlas is compiled by COZLABS, a small independent group, with extensive use of large language models for locating passages, drafting entries and notes, and writing the software that builds the site and the maps. The rules in this note exist largely to keep that work honest: quotations are checked against a base text, identifications carry their grade, nothing is placed or linked without a stated basis, and a script counts certain habits of machine-written prose in the text readers see so that they do not creep back. Errors remain, and some are listed on the doubts page and in the journal. Corrections are welcome through the letter form on the English page.

12. Known weaknesses

Many texts are read in digital editions rather than in printed critical editions, and the digital text is sometimes an annotated selection (the Haiguo tuzhi) rather than the full work. Much of the modern scholarship is known through secondary summaries, and is graded accordingly. Identifications of places in the Western Regions and India depend heavily on the DILA Buddhist place authority database and the modern commentary it cites. The houses on West Lake rest on local history, the press and family recollection, with few archival documents. The count of names on Ricci's map (964) differs from published counts (1,114; nearly 1,200) because the two hemisphere insets and the astronomical diagrams are not counted.

13. Citation

The current web edition includes later corrections to map references and the shared datasets. The DOI below identifies the archived method note, version 1.1, dated 29 September 2026. For this edition, cite https://yutuji.com/en/method/ with the access date.

COZLABS. Yutuji: a hand-drawn historical atlas with graded sources. Method note, version 1.1. Zenodo, 2026. https://doi.org/10.5281/zenodo.23019771

COZLABS. Yutuji Historical Atlas: Open Data. Zenodo, 2026. https://doi.org/10.5281/zenodo.23010865