A Western reader hits the phrase "turned agent" in a Chinese military text and thinks he knows what it means. He pictures a recruited spy who has changed sides — a man who once served the enemy and now, by choice, serves you, feeding his old masters false reports while reporting truly to his new ones. That picture is wrong often enough that Sawyer makes a point of warning about it, and the warning is one of the load-bearing arguments of his whole book.1
The Chinese fan chien — what gets translated "turned agent" or "double agent" — usually means something narrower and stranger to the Western ear: an enemy spy who was detected and then converted or bribed, or an enemy insider who is being manipulated and exploited without ever changing allegiance at all, without even knowing he is being used.2 The Western double agent requires a conscious, if not always willing, choice. The Chinese concept often requires no choice from the agent whatsoever. The gap between the two is not a translation nuisance to be smoothed over. It is an epistemological problem about what we are even able to recognize when we read across the two traditions.
Sun-tzu's double agent was, at root, an enemy spy who had come to spy on you, been caught, and been "tempted with profits, instructed and retained."3 But the concept did not hold still. Sawyer flags the drift in a parenthetical that is doing heavy lifting: Sun-tzu's definition "eventually evolved to include enemy agents who were being unconsciously manipulated."4 And here is the crux — "most traditions would not term such individuals 'double agents' or 'turned agents' given the absence of any volitional component. However, throughout Chinese history the process of converting or turning them to one's own use was the defining factor, not whether they were knowledgeable or participants."5
Sit with that. In the Western frame, the agent's decision defines the category — a double agent is someone who decided to deceive his own side. In the Chinese frame, the handler's action defines the category — a turned agent is someone the handler has turned to his own use, whether the agent agreed, noticed, or did anything at all. The volition lives in a different person. Sawyer notes the term is "often employed in later commentaries discussing what should be simply 'living spies,'" which means the Chinese sources themselves used fan chien loosely, blurring the manipulated insider into the genuinely converted agent.6
Translation could solve a vocabulary problem. You'd pick a better English word and move on. This is worse, because the categories carve reality at different joints. A Western analyst reading "turned agent" imports a whole apparatus of assumptions — the agent knows, the agent chose, the agent could in principle re-defect, the agent's loyalty is the thing at stake and the thing to be doubted. None of that need be true of a fan chien who is simply an enemy courier being fed false information he believes is real. There is nothing to re-defect from; he never defected. His "loyalty" was never engaged; only his usability was.
This is why Sawyer keeps a translation note running across the book — the same caution surfaces again far later, where he distinguishes "false double agents" and "expendable agents planted to spread disinformation by feigning a new loyalty" from genuine double agents, and recommends comparing a double agent's reports against other sources before trusting them.7 The recurring note is the tell: the concept is slippery enough that the translator has to keep re-stabilizing it. When a category needs constant re-explanation, the category is not transferring cleanly between traditions — and the reader who forgets that will systematically mis-read the Chinese record, projecting a volition that isn't there.
This page is a correction lens for the whole intelligence corpus. Any vault page that uses "double agent" or "turned agent" of a Chinese source should pass through here first, because the phrase may name a manipulated insider rather than a willing turncoat. It supplies a single diagnostic question — whose volition defines this category, the agent's or the handler's? — that prevents the corpus from importing a Western mental model wholesale into a tradition that organized the same terrain differently.
It also gives the vault a model case of how concepts fail to translate. The lesson generalizes past spycraft: any time a tradition's key term gets rendered with a familiar Western word, the familiarity may be smuggling in assumptions the original never held.
Sawyer is careful to say genuine double agents did exist in China, and he gives one from the Three Kingdoms period precisely because it is the Western kind, the exception against which the usual blur stands out. Wei-ch'ü — "hard, fierce, astute, and cruel," so feared that all his subordinates trembled — sent two of his close confidants to gather information on the regional governor Chang Ni under the pretext of wanting to surrender.8 Chang Ni "realized their true objectives, and by granting generous rewards converted them to double agents," then plotted with them to assassinate Wei-ch'ü. When it was done, Wei-ch'ü's band "all felt secure."9
This is the Western pattern — detected enemy agents, consciously converted by reward, knowingly working against their original master, even to the point of murdering him. So why does Sawyer present it as notable rather than typical? Because its very clarity is unusual. He has to reach to the Three Kingdoms for a clean case, and he frames it explicitly against the broader Chinese tendency to call manipulated-but-unconverted insiders "turned agents." The case proves the rule by contrast: the genuine volitional double agent is real but is the marked case, the one worth singling out, against a background where fan chien usually meant something with no volition in it.
Set beside it the freelance operator Su-tzu, who is the opposite extreme — pure volition, no allegiance at all. When Western Chou dammed the river to ruin Eastern Chou's rice, Su-tzu volunteered to go to Western Chou and argued them into releasing the water — on the cynical logic that flowing water would let Eastern Chou plant rice they could later be denied again, keeping them dependent.10 "Su-tzu thus received payment from two states."11 He served no one and sold both — a freelance triple-agent extracting water and gold from both Chous at once. Wei-ch'ü's confidants and Su-tzu bracket the spectrum: the converted-by-reward turncoat at one end, the loyal-to-no-one mercenary at the other, and in the wide middle the manipulated insider whom the Chinese tradition calls "turned" though he never turned anything.
The analyst has a Chinese campaign record open and the phrase fan chien under his finger. The reflex is to read "double agent" and picture the man who chose. He stops himself and asks the diagnostic question instead: did this person decide anything? He reads on. The "turned agent" here was an enemy courier captured and fed a false dispatch he carried back believing it true — no choice, no change of heart, no loyalty in play, only a man used like a tool and never the wiser. The analyst writes "manipulated conduit," not "double agent," in his notes. He has just avoided importing an entire false model — the assumption that this man could be flipped back, that his loyalty was the variable, that he knew. Next entry, different story: an enemy spy, detected, paid, knowingly assassinating his old master. This one chose. Here "double agent" is right. The discipline is to make the same phrase earn the Western word every single time, and to deny it that word by default.
You are failing this concept when "double agent" passes your eye without friction. The diagnostic sign is unexamined familiarity: the word looks like one you know, so you stop reading carefully. A second sign is attributing motive to a manipulated insider — narrating his "decision to betray" when the sources record no decision at all, only a handler's exploitation. A third is treating the Three Kingdoms clean case as the norm rather than the marked exception, and so over-counting volitional turncoats in a record that mostly contains used men.
The deepest failure is forgetting that the gap is the data. Smooth the two concepts into one and you have not translated the Chinese tradition; you have overwritten it with your own.
The central tension sits inside the Chinese sources themselves, not just between traditions. The commentaries use fan chien loosely — "often employed...discussing what should be simply 'living spies'" — which means the blur is native, not merely a Western misreading.12 So there are two layers of imprecision stacked: the Chinese tradition was already loose about the term, and then Western translation adds a second distortion by importing volition. Sawyer documents both layers without claiming to resolve either, which is the honest posture; the looseness is a real feature of the historical material.
An open question Sawyer leaves standing: if the defining factor was the handler's act of turning rather than the agent's knowledge, did Chinese practice actually distinguish the witting from the unwitting agent operationally — handling them differently in the field — even while collapsing them terminologically? The text suggests the distinction mattered in practice (you'd hardly run a knowing assassin the way you'd run a deceived courier) but was not consistently marked in the vocabulary. Whether that is a gap in the sources or a gap in the concept is not settled.
Sawyer the comparativist is the whole engine of this page, and his project has a built-in strain worth naming. To make the Chinese concept legible he must describe it in contrast to "the contemporary Western idea of a double agent" — yet the contrast risks the very thing he warns against, letting the Western category set the terms by which the Chinese one is judged "incomplete" (lacking volition). His own framing — that volition is "absent" — quietly treats the Western definition as the baseline from which the Chinese one deviates. A different telling would make the Chinese definition the baseline and call the Western fixation on the agent's choice the parochial one. Sawyer's cross-period sweep (Spring and Autumn couriers, a Three Kingdoms conversion, Warring States freelancers) again asks the reader to trust a continuity across many centuries; the continuity is plausible but is the translator's synthesis, and the looseness of fan chien across that span may reflect changing usage rather than one stable concept.
Handshake one — The Semblances Problem. The semblances problem is the ancient difficulty that an appearance and its opposite can look identical — the loyal man and the traitor present the same face. The turned-agent confusion is a special, sharpened case: here the category itself is a semblance. "Turned agent" looks like "double agent," and the resemblance of the words hides a difference in the things. The insight that neither domain reaches alone: the semblances problem usually operates on people (which man is the spy?); the turned-agent case shows it also operates on concepts (which meaning is the word carrying?). A handler can be deceived by a face; an analyst can be deceived by a translation. The epistemology of mistaken resemblance runs all the way up from the agent in the field to the scholar at the desk — the same failure of discrimination, one level higher.
Handshake two — Su Ch'in the Counterfeit Double Agent. Su Ch'in is the vault's case of an operator who manufactured the appearance of being a double agent — a counterfeit. Read against this page, a three-way distinction the corpus needs comes into focus: the genuine volitional turncoat (Wei-ch'ü's confidants), the manipulated insider with no volition (the usual fan chien), and the counterfeit double agent who only performs the role (Su Ch'in). The insight: "double agent" in the Chinese record is not one category but at least three, separable only by asking where the volition and the deception actually sit — in the agent, in the handler, or staged for an audience. Su Ch'in's counterfeit and the unwitting conduit are opposite errors a careless reader makes with the same phrase: one over-attributes performance, the other over-attributes choice.
Sharpest implication. When a foreign tradition's key term gets translated with a word you already know, the familiarity is the danger. "Turned agent" reads as "double agent" and thereby smuggles in a volition the Chinese concept often never had — and the analyst who doesn't catch the smuggling will misread the entire record in a consistent, invisible direction.
Generative questions.