Terracotta Army fingerprints: who did the work?

The names stamped on China's Terracotta Army belong to a few dozen accountable craftsmen. The fingerprints in the clay tell a different story. A 2026 study of 79 prints estimated that about 45% came from adult men, 33% from adult women and 22% from children under 15.

That gap between whose name is on the work and whose hands are in it is a very old problem, and performance reviews still have it. Before using it, though, check what the number actually measures. It's less than the headlines suggest.

79 prints and what they can tell us

The study counted fingerprints. It didn't count workers, and everything else follows from that.

The paper is "Who produced Emperor Qin Shihuang's Terracotta Army, China? A Bayesian mixture model approach to fingerprint age and sex determination" by Xiaoxi Li and colleagues at the Emperor Qinshihuang's Mausoleum Site Museum, published in the Journal of Cultural Heritage on October 13, 2026. It's an observational study. You measure prints left on artifacts, then fit a statistical model. I couldn't open the paper itself (the publisher blocks automated access), so the details below come from cross-checking ScienceAlert (October 17, 2026), the South China Morning Post (October 21) and Ancient Origins.

The sample: 36 warriors in one section of Pit 1, labeled T23. The team documented 91 prints and dropped 12 that seemed to come from people who had already left other prints. That left 79: 73 pressed into the clay surface, mostly on heads and arms, and 6 relief prints on armor nails and waistbands.

The measurement is mean ridge breadth, the average spacing between fingerprint ridges. Spacing widens as you grow, stops at skeletal maturity, and is slightly wider in adult men than adult women. The team did not label each print "woman" or "child." A Bayesian mixture model estimated which mix of people would most likely produce the whole spread of 79 measurements, after correcting for clay shrinkage during drying and firing.

The estimate: 45% adult men, 32.7% adult women, 22.3% under 15. Most coverage rounds the last two to 33% and 22%.

It doesn't mean "most of the army was made by women and children"

Prints are a biased trace, and the estimate has wide error. The researchers themselves say they may be overstating the role of women and children.

Skilled craftsmen likely used tools and smoothed their marks away. Beginners leave more prints. Removing likely duplicates helps, but it can't guarantee that one busy pair of hands isn't counted several times. So 22% of prints is not 22% of the workforce.

The method has a ceiling too. A 2023 methods paper by Burchill, Sanders and Morgan in MethodsX, which develops this Bayesian approach for ancient potters, states that expected values for adult women and adolescent boys are "inherently indistinguishable." The model carries that overlap as uncertainty. It can't remove it. And the reference data come from modern European populations. Nobody has an equivalent set for people in Qin China 2,200 years ago.

Then the denominator. Smithsonian Magazine reported in 2020 that a little over 2,000 figures had been excavated, out of an estimated 7,000. Britannica puts the army at around 8,000. This study looks at 36 of them, from one spot.

What survives all that is modest. The long-held assumption that only trained adult men made the army doesn't fit the physical evidence, at least in this section. Qin bamboo-slip records already described apprentices and women and children in state workshops. Now there are hands to go with the text. I'd stop there.

What the stamped names were for

Accountability. Credit had little to do with it. Classical texts record the rule as 物勒工名 ("the object is inscribed with the maker's name, to test his sincerity"), and Qin law made it a system. If a piece failed, you knew whose it was.

A 2016 paper by Xiuzhen Li, Andrew Bevan and colleagues in the Journal of Anthropological Archaeology mapped the names on the warriors. Ten personal names following a "Gong" (palace workshop) stamp appear on 55 figures. Eighteen names linked to "Xianyang" appear on 49. The authors suggest each named artisan may have overseen a small group of less skilled collaborators.

Put the two studies side by side. The names are the people who answered for the figure. Many of the hands that pressed the clay and set the armor nails left no name anywhere.

The org chart version

From here on, the leap is mine. A Qin workshop is not a company, and child labor is not an analogy for interns. But the structure looks familiar to me.

One accountable name per deliverable is good design. I've argued for one owner per module because shared ownership quietly dilutes effort. The trouble starts when the review process reads the owner field as if it were the contribution field.

The fingerprints add a second twist, because the bias runs both ways. Juniors leave visible traces, like messages, tickets and commits. Seniors smooth theirs away. They fix other people's work, catch the incident before it happens, rewrite the deck the night before. Count names and you over-credit owners. Count activity and you over-credit the noisy. Neither is a headcount of who did the work. That's part of why the people who never appear on the org chart are so easy to lose, and so expensive when they leave.

For foreign managers in Japan this matters more than at home. Japanese teams tend to keep individual contribution low-key, and the person presenting upward is often not the person who did the analysis. If HQ only sees the presenter, it is reading stamps.

The practical fix is cheap. Keep the single owner. Next to it, keep a short record per deliverable of who touched what. In review conversations, ask the owner once: whose hands are on this besides yours? And when you interview, ask for one specific case. I once interviewed a head-of-sales-strategy candidate who listed a long run of initiatives and stalled every time I asked about a specific one. His name was on a lot of work. I couldn't find his prints.

The fingerprint numbers may move as the team samples more pits and builds better reference data, which they say they need. What I'd keep is the habit of counting names and hands separately.