The Chinese intelligence tradition understood something counterintuitive about human recruitment: the qualities that make a person an excellent spy for one function make them actively dangerous for another. A surviving spy who is too courageous becomes reckless. A doomed spy who is too intelligent will realize what he actually is and the operation collapses. A converted spy who is genuinely loyal to the enemy remains exactly that. The mission determines the person; the person does not determine the mission. This is the deep logic of the five-spy typology — it is not a taxonomy of espionage functions but a framework for understanding that each function requires a distinct character profile, and that selecting the wrong profile is worse than having no agent at all.
The tradition's most demanding claim: the handler's character determines the network's effectiveness more than the agents' characters. Sun Tzu's Chapter XIII argument is precise — foreknowledge requires "the sovereign's personal sagacity" to interpret and act on intelligence correctly. A brilliant agent reporting to a handler who cannot assess men or situations produces wasted intelligence. The network is only as good as its organizing intelligence, not its executing members.
Local spies (hsiang-chien) — recruited from the enemy's civilian population. The qualification requirement here is not courage or tradecraft but embeddedness: local spies must have genuine terrain knowledge (routes, water sources, defensible ground, local power structures) and genuine social networks that provide natural cover for their movement and conversation. They are not trained operatives — they are existing nodes in a social network who are converted into information channels. Selection criterion: natural authority or social connection within the target population. What you cannot manufacture is what makes them valuable.
Inward spies (nei-chien) — recruited from the enemy's own officials. The qualification requirement is a compound of three elements that rarely appear together: sufficient access to possess information worth gathering, sufficient dissatisfaction with the existing regime to accept the approach, and sufficient reliability under pressure to not betray the operation when the stakes rise. These three requirements frequently contradict each other — the official with greatest access may have the most to lose from betrayal and therefore the highest threshold for recruitment; the most dissatisfied official may be too emotionally unstable to be reliable. The selection problem for inward spies is finding the intersection of a Venn diagram with three rarely-overlapping circles.
Converted spies (fan-chien) — turned enemy intelligence operatives. The tradition's most complex qualification analysis: the converted spy is simultaneously the most valuable and most dangerous type. Most valuable because they come with existing access, training, and knowledge of the enemy's intelligence infrastructure. Most dangerous because their loyalties are structurally unverifiable — a converted spy who appears to have been turned may still be running the original operation, now with elevated access to the handler's network. The qualification requirement is paradoxical: you are selecting for someone who can be genuinely turned, which is itself evidence of a character structure that cannot be fully trusted. Li Ch'uan's T'ang dynasty commentary on this type focuses on the handler's responsibility — the converted spy must be "treated with the utmost liberality" not from sentiment but from strategic calculation: the handler who creates strong material and personal bonds with the converted spy is actively working against the pull of the spy's original loyalties. The qualification for a converted spy is as much a qualification of the handler's capacity to maintain the bond as it is a qualification of the spy's character.
Doomed spies (ssu-chien, also translated as expendable spies) — agents deployed to carry false intelligence into enemy territory, knowing they will likely be captured and killed when the intelligence is discovered to be false. The doctrine contains an ethical provision that is the most disturbing element in the entire tradition: the doomed spy must not know they are doomed. The doomed spy must believe the intelligence they are carrying is genuine — because if they know it is false, they will be unable to convince the enemy interrogator with the sincerity required for the operation to work. The handler deliberately creates a false information environment for their own agent. The qualification requirements follow from this design: the doomed spy must be courageous enough to function under capture and interrogation, credible enough for the enemy to believe their report, and — critically — not intelligent enough to penetrate the false information environment they have been placed in. Selecting a doomed spy is selecting for the absence of a quality (strategic insight) that is prized in every other type.
Surviving spies (sheng-chien) — long-term intelligence assets who penetrate the enemy's apparatus and return with information. The qualification requirements here are the most comprehensive: tradecraft (the ability to move, communicate, and maintain cover without triggering observation), reliability over extended periods under sustained pressure, and the specific psychological quality of enduring long-term role performance without losing operational judgment. The surviving spy is performing a false identity continuously, sometimes for years. The selection criterion the tradition emphasizes above others is route — the surviving spy must have a reliable exfiltration path before they are deployed, because an agent who cannot return safely is a future doomed spy, not a surviving one.
Li Ch'uan's T'ang dynasty commentary adds a type not found in Sun Tzu's original taxonomy: the roving agent (hsing-chien — the moving, circulating, or visiting agent). These are agents without fixed position who move between territories, courts, and factions — intelligence gatherers whose cover is mercantile, diplomatic, or scholarly travel. Their qualification profile differs from all five Sun Tzu types: they require broad social range rather than depth of access (the inward spy's access is to specific fixed officials; the roving agent's access is to any network they move through), adaptability across social contexts rather than sustained performance of a single cover identity, and the specific intelligence of recognizing what matters in an unfamiliar environment.
Li Ch'uan's roving agent addresses a gap in the original five-type typology: Sun Tzu's five types all assume the agent is either already embedded in or specifically inserted into a fixed target. The roving agent is the intelligence solution for dynamic, uncertain, or rapidly changing situations where no fixed embedding is possible. The qualification for the roving type is the capacity for rapid pattern recognition in novel social environments — which is almost the opposite of what makes the inward spy effective.
Sun Tzu's Chapter XIII makes an argument that the qualifications discussion tends to obscure: the network's effectiveness is a function of the handler's sagacity, not primarily the agents' qualities. "Unless the sovereign is enlightened and wise, he cannot use spies. Unless he is benevolent and just, he cannot employ spies. Unless he is subtle and perspicacious, he cannot perceive the truth of intelligence reports."
This is not a platitude about virtue — it is a structural claim about information processing. An intelligence network generates data. The value of that data depends on the handler's ability to assess which data is reliable, which is planted, which is incomplete, and what it implies for action. The most qualified agents reporting to a handler who cannot assess their reports are generating noise, not intelligence. The handler's qualification for the chih jen tradition (knowing men, reading character) is upstream of every agent qualification — because without it, the entire output of the network is uninterpretable.
The tradition is explicit about what each agent type should and should not know. Doomed spies must not know the false nature of their intelligence — already established. But the compartmentalization extends further: agent types should be unknown to each other. The surviving spy in the enemy court should not know that a converted spy has also been placed there. The local spy providing terrain intelligence should not know who the handler's inward source in the enemy government is. The principle is not just operational security against capture — it is security against the structurally difficult converted spy. If the converted spy knows the full network, turning them back provides the enemy with the full network.
Compartmentalization is the structural answer to the converted spy's dual-loyalty problem: you cannot fully verify the converted spy's loyalty, so you architect the network so that the maximum damage the converted spy can cause is bounded by what they actually know.
Sun Tzu's original five-type framework and Li Ch'uan's T'ang dynasty additions are not in tension so much as at different scales of concern. Sun Tzu's framework is mission-typological — it classifies agents by their relationship to the target (local/inside/enemy-turned/sacrificed/surviving). Li Ch'uan's additions, including the roving agent, are functionally typological — they classify agents by the kind of intelligence work they do. These are complementary rather than competing, but the shift in classificatory principle reveals something about what each thinker is optimizing for. Sun Tzu is optimizing for the recruitment relationship: how does the handler obtain agents with access? Li Ch'uan is optimizing for the intelligence product: what kind of information does the handler need, and what operational profile produces it? The T'ang dynasty additions represent an empirical expansion based on two centuries of documented intelligence practice since the Sun Tzu tradition was codified — and the roving agent type specifically suggests that the fluid political landscape of Tang China created intelligence requirements that the five-type model couldn't cover.
The handler-quality argument in Sun Tzu Chapter XIII sits uneasily with the agent-qualification framework as a whole. If the sovereign's sagacity is the primary determinant of network effectiveness, then the detailed qualifications for each agent type are secondary — and the tradition's extensive typological analysis of agents might be directing attention at the wrong variable. The qualification that matters most is the handler's qualification, and the tradition discusses this least — perhaps because it was unpolitic to tell rulers that the limiting factor in their intelligence apparatus was themselves.
The mission-specific character-matching logic of agent selection — different functions require fundamentally incompatible character profiles — appears in two other domains where the same design problem operates with different vocabularies.
Cross-Domain: The Semblances Problem — the expendable agent doctrine is the semblances problem deliberately engineered against the handler's own operative. The doomed spy who believes false intelligence is genuine is not encountering a naturally occurring semblance — the handler has manufactured one specifically for them, with full knowledge of its falsity. This is the most extreme application of Chieh Hsüan's obfuscation (noise) framework: noise is being used not to deceive an enemy but to make one's own agent into an unknowing instrument of deception. The cross-domain insight: the semblances problem is usually framed as a threat to the assessor — how do you distinguish genuine from manufactured signals? The doomed spy doctrine reveals that the same problem can be deliberately installed in an ally to weaponize them. Manufactured belief produces authentic behavior; that is the mechanism in both cases.
Psychology: Character as Procedural Learning — Scaer's account of procedural memory and character encoding reframes what the agent-qualification tradition is actually selecting for. When the tradition specifies that the doomed spy should not be "too intelligent," it is identifying a specific procedural encoding — the capacity for strategic meta-cognition — that would interfere with the operative's function. When it specifies that the surviving spy must be able to sustain long-term role performance, it is identifying a procedural encoding of emotional regulation and identity flexibility that most people do not possess. The cross-domain insight: agent qualification is not a character-trait checklist but a search for specific procedural encodings that the mission requires and that cannot be trained in the short term. The tradition understood intuitively what Scaer's framework makes explicit: the procedural layer of character is what matters for performance under pressure, and it is encoded through history rather than selected through will.
The Sharpest Implication
The doomed spy doctrine is the most ethically naked statement in the entire Chinese intelligence tradition — and what it implies, if taken seriously, is that any sufficiently complex intelligence operation will contain within it a sub-population who are being systematically deceived by their own handlers about the nature of what they are doing. This is not a failure or corruption of the system; it is the system working as designed. The doomed spy's ignorance of their status is not a gap in the handler's communication — it is the primary operational requirement. If you accept the logic of the doomed spy doctrine, you accept that the handler's duty to the operation supersedes the handler's duty to the operative's informed consent. The same reasoning — that the mission's requirements can override the operative's knowledge of their own situation — appears in any sufficiently hierarchical organization where operational compartmentalization is strict. The doomed spy is the limit case, but the principle is not exotic.
Generative Questions