The previous lesson settled on a deliberately broad definition of the subject and warned that the only thing keeping it honest is evidence, so this lesson looks at what that evidence actually is, and it is stranger and thinner than most accounts admit.
The sample nobody took
Start with what a historian would want. To say how much trade crossed Eurasia, you would want customs records from a representative set of crossing points across several centuries, the way an economic historian of eighteenth century Britain has port books and excise returns. Nothing of the kind exists for any part of this subject, at any date, anywhere.
What exists instead is a scatter of documents that survived because of where they were dropped. There is no sampling frame, no way of estimating what fraction of the original writing has come down, and no reason to think that fraction is even roughly constant across regions. A historian of the Silk Road is in the position of an archaeologist trying to describe a city from the contents of four rubbish pits, three of which are in the same street.
That is not a counsel of despair, and the material that does survive is extraordinarily rich. It is a warning about a specific and very common error: reading the density of evidence as the density of activity. The Tarim Basin has produced more everyday documents from the first millennium than the whole of northern India, and the reason has nothing to do with how much happened in either place.
Why the desert keeps things
Organic material survives when the microbes that eat it cannot work, and they need water. Annual rainfall at Turfan is around 16 mm, at Dunhuang around 40 mm, and evaporation exceeds it by orders of magnitude. Paper, wood, silk, leather, grain, human bodies and their clothing sit in that sand for two thousand years and come out legible.
Compare the alternatives. In the Ganges plain, in Sogdiana's irrigated lowlands, in the Chinese south, in the Mediterranean outside Egypt, the same objects are gone within a few centuries. The palm-leaf manuscripts of Indian Buddhism are known largely from copies made in Nepal and Tibet, or from fragments found in Central Asia, which is to say that even Indian religious texts reach us through the desert. Sogdian, the language of the merchants who ran the trade for six hundred years, is known mainly from documents found outside Sogdiana.
Set out plainly, the asymmetry is severe: the places we can see are the dry, marginal, thinly populated ones, and the places we cannot see are the wet, rich, populous ones. Any statement of the form "most of the trade was in the Tarim" is therefore suspect on its face, because the Tarim is where the evidence is.
Example. An excavation report notes that roughly seven hundred documents survive from Niya, a small oasis in the southern Taklamakan abandoned around 400 CE, and that essentially nothing comparable survives from Pataliputra, the great city on the Ganges. What may we conclude about their relative commercial importance?
Nothing at all, and the temptation to conclude something is exactly the error this lesson exists to prevent. The two sites differ in the one variable that controls document survival, which is water, and they differ in it so extremely that the comparison carries no information about anything else. Pataliputra sits in a monsoon flood plain where wood and birch bark and palm leaf decay within generations, and it has also been continuously occupied, which means later building has destroyed earlier layers. Niya was abandoned and never reoccupied, and it is bone dry. To compare the commercial importance of the two you would need a class of evidence that survives equally in both, such as coins, imported ceramics or building remains, and if you use those the ranking reverses at once.
Now you. A historian argues that Buddhism was more important in Central Asia than in India during the fifth century, citing the enormous quantity of Buddhist manuscripts recovered from Central Asian sites and the near absence of contemporary Indian ones. Diagnose the argument.
Answer
It is the same error with a different subject. Manuscript survival is controlled by climate and by whether a site was abandoned, so the comparison measures aridity rather than devotion. The correction is to ask what evidence would survive equally in both places, and for Buddhism there is a good deal of it: monastic architecture, stupas, sculpture, cave complexes, royal inscriptions and land grants on copper plates and stone, all of which survive in India in quantity and show a fifth century Buddhist establishment that was large, endowed and politically connected. There is also a neat internal check, which is that the Central Asian manuscripts are overwhelmingly translations or copies of texts composed in India, and a translation is evidence of a source. The honest version of the claim is narrower and still interesting: Central Asia is where we can watch Buddhism being transmitted, because the transmission left paper and the origin did not.
Four pits
Four bodies of material carry most of the weight, and it is worth knowing them by name, because nearly every specific claim in the rest of this course descends from one of them.
The Niya documents were found by Aurel Stein from 1901 onwards at a site in the southern Taklamakan, capital of a small kingdom the Chinese called Shanshan. There are roughly seven hundred of them, written in Gandhari Prakrit in the Kharosthi script, an Indian language and an Indian script in a town two thousand kilometres from India, which is itself a finding. Many are wedge-shaped wooden double tablets, tied with cord and sealed with clay so that the text inside could not be read without breaking the seal, an envelope made of wood. Their contents are relentlessly local: the sale of a camel, the ownership of a vineyard, a dispute over a runaway slave, an order from the king about monks behaving badly. There is almost no international trade in them at all.
The Sogdian Ancient Letters, five of them substantially complete, were found by Stein in 1907 in a ruined watchtower on the Han wall west of Dunhuang. They are the oldest substantial Sogdian texts known, and they are business and family correspondence written on paper. The reason they are in a watchtower and not in Samarkand is that they were never delivered. A mailbag, abandoned or confiscated, is the single most informative kind of document a historian can find, because it has not been selected for importance by anybody.
The Turfan documents come from the Astana cemetery and other sites in the Turfan oasis, and reach us through a bizarre route: waste paper was recycled into burial goods, so that shoes, hats and figurines for the dead were made from discarded contracts, tax registers, market price lists and school exercises. Peeling apart a paper shoe from a seventh century grave yields the everyday administration of an oasis town.
The Dunhuang library cave, cave 17 of the Mogao complex, was walled up early in the eleventh century and opened in 1900 by a Daoist caretaker named Wang Yuanlu. It held something like fifty thousand manuscripts and printed items, in Chinese, Tibetan, Sogdian, Khotanese, Uighur, Sanskrit and more. Among them is the Diamond Sutra printed on 11 May 868, the earliest complete printed book bearing its own date. Why the cave was sealed is unresolved: the two live proposals are that it was a store of sacred waste, texts too holy to burn and too damaged to use, and that it was hidden in a hurry against an expected invasion. Rong Xinjiang's argument for the sacred waste reading rests on the state of the contents, which are heavily weighted towards damaged and duplicate material rather than the treasures a monastery would evacuate.
Example. Of the roughly fifty thousand items from cave 17, the overwhelming majority are Buddhist scripture, and the secular material, contracts, loan agreements, household registers, lay society circulars, is usually put at a few per cent. Take five per cent as a working figure and say what follows.
Five per cent of fifty thousand is 2,500 documents. That number is the entire documentary base for the social and economic history of one oasis over roughly six centuries, and it is by a wide margin the largest such collection in Central Asia. Two consequences follow. The first is that any statistical claim built on it, average loan interest, the frequency of a contract type, the price of a bolt of cloth, rests on tens or low hundreds of instances once the material is sorted by century and category, which is why the honest literature quotes ranges rather than means. The second is that Dunhuang inevitably becomes the default picture of the whole region, simply because nowhere else can answer the question at all, and a reader should be alert whenever a general statement about Central Asian society turns out to be a statement about one town.
Now you. Suppose the sacred waste explanation for the sealing of cave 17 is correct. How should that change the way you use the collection as evidence?
Answer
It changes it in a useful direction, because sacred waste is closer to a random sample of what the monastery had than a deliberate deposit would be. If the cave had been packed against an invasion, the contents would be what someone in the year 1000 judged worth saving, a strong and unknowable selection filter. If it is discard, the filter is different and much weaker: damaged, superseded and duplicate material, plus whatever scrap paper happened to be swept up with it, which is precisely why so much secular administration is in there at all. Nobody preserved a loan contract because it mattered. The catch is that discard is not unbiased either, since it over-represents what a monastery accumulated in bulk and under-represents anything valuable enough to be repaired or kept in use, and it says nothing about lay households that had no monastery to leave their paper in. So the collection is better evidence than a hoard would be, and still evidence about one institution.
Dating something nobody dated
Most of this material carries no date, and the techniques used to supply one are worth seeing once, because they decide how much weight a document can bear.
The Sogdian Ancient Letters are the standard case. Letter II describes a catastrophe in China: the last emperor, the writer says, fled Luoyang because of famine, his palace was set on fire, and the city and Ye are no more. It also reports that Indians and Sogdians in Luoyang had all died of starvation. Chinese sources record the sack of Luoyang by Xiongnu forces in 311 CE and the flight of the court, and W. B. Henning argued in 1948 that the letter must have been written shortly after, dating the group to 313 or 314. János Harmatta later proposed 196 CE, matching the letter to an earlier round of disorder at the end of the Han.
The dispute is settled, for most scholars, on the details rather than the general fit: the sequence of events in Letter II matches 311 closely, and the letters' Sogdian is late enough in form to sit uncomfortably in the second century. What matters here is the shape of the reasoning. The letters are dated by tying an internal narrative to an externally dated event, which means the date is only as good as the match, and the whole chronology of early Sogdian trade rests on one letter's account of a fire.
Evidence that is not writing
The subject has been transformed since about 1990 by evidence that no scribe produced. Coin finds map circulation independently of any text. Isotope analysis of human teeth shows where a person spent childhood, which is how cemeteries of apparent locals have turned out to contain first generation migrants. Textile analysis identifies where a cloth was woven from its weave structure and its fibre: the silks excavated at Palmyra in the 1930s include Han damasks whose patterns and technique are Chinese, and that identification is a laboratory result rather than a reading of Pliny.
Example. A silk fragment from a Palmyra tomb is identified as Chinese from its weave structure rather than from its pattern. Why is that the stronger form of identification?
Because a pattern can be copied and a loom setup cannot be, at least not without building the loom. Motifs travel: a Syrian weaver who has seen an imported Chinese textile can imitate its dragons or its cloud scrolls on local equipment, so a Chinese-looking design proves only that somebody had seen Chinese design. The structure of a warp-faced compound weave, by contrast, records the physical arrangement of the loom that made it, the number of warp sets, how the sheds were controlled, the twist and preparation of the thread. Reproducing it requires reproducing the whole apparatus and the skill to work it, which is the point the tenth lesson develops at length. So structure is close to a fingerprint of a manufacturing tradition, and it is the kind of evidence a laboratory can assess without any historical assumption at all. The general principle is worth keeping: prefer the evidence that is hard to imitate over the evidence that is merely characteristic.
Now you. The Niya documents carry no calendar dates that a modern reader can convert directly. How would you date them?
Answer
By tying them to something else, using as many independent lines as you can, which is the general method. The documents are internally dated by the regnal years of local kings, so you can build a relative sequence of reigns and then anchor it: several of those kings appear in Chinese sources, whose chronology is absolute, which fixes the whole series. Objects in the same layers help, particularly Chinese coins, whose issue dates are known, and imported goods whose typologies are dated elsewhere. The wooden tablets themselves can be radiocarbon dated, with the caveat that this dates the death of the tree rather than the writing, and wood in a treeless oasis was reused. Palaeography, the changing shapes of Kharosthi letters, gives a rough independent check. Notice that every method here works by connecting the undated material to a dated system somewhere else, which is why regions with no literate dating neighbour are so much harder to work on, and why the Sogdian letters had to be dated by a fire in Luoyang.
The most dramatic recent case is ancient DNA. A cemetery in the Chu valley of Kyrgyzstan, excavated in the 1880s, has tombstones dated 1338 and 1339 that record deaths from pestilence; in 2022 a team led by Maria Spyrou published Yersinia pestis genomes recovered from teeth in those graves, placing the strain at the root of the lineage that caused the Black Death. That is a chain of evidence, gravestone date to genome to phylogeny, in which no historical text appears at any point, and a later lesson is built on it.
How to read a claim
The working habit this lesson is trying to install is a question to ask of any specific statement about the Silk Road: what is this resting on, where was it found, and how was it dated?
Try it on three claims. "Sogdian merchants had commercial networks reaching from Samarkand to Chang'an" rests on the Ancient Letters plus Sogdian colonies attested in Chinese administrative sources plus tomb reliefs in China, three independent classes of evidence, and it is about as secure as anything in the field. "The Silk Road carried enormous volumes of trade" rests, when you chase it, mostly on Roman moral complaints about luxury and on the presence of exotic objects in elite tombs, neither of which measures volume. "Paper making passed to the west after the battle of Talas in 751" rests on a single anecdote written down nearly three centuries later, and the next chapters of this course take it apart.
The next lesson changes tack and asks a question that needs no documents at all. Before asking what moved, ask what could move: a loaded camel has a range, a payload and an appetite, and those three numbers constrain the trade more tightly than any political history does.