The Circle Framework

The few are always weaker than the many. Hume drew the consequence in 1741: since force is always on the side of the governed, the governors have nothing to support them but opinion, and every government, the most despotic and the most free alike, rests on opinion alone ("Of the First Principles of Government"). La Boétie had said it two centuries earlier from the other side: a tyrant is not overthrown by fighting him but by ceasing to hold him up, and he stands only because the many consent to stand under him (Discours de la servitude volontaire, c. 1576). Weber turned the observation into a definition. Domination is not obedience obtained; it is obedience obtained because the subject believes the command to be legitimate, and legitimacy, whatever its type, is nothing other than that belief (Economy and Society, 1922).

Power, then, is a belief system: a network of associations in the head of each subject that links the ruler to rightness, the order to necessity and dissent to danger. It has the same mechanics as any other belief system, and it can be built, moved and defended by the same means.

Lukes gives the distinction that locates where the work is done (Power: A Radical View, 1974). Power has three dimensions. The first is winning a visible conflict: an election, a vote, a fight. The second is keeping the conflict from occurring by controlling what reaches the agenda. The third is shaping what people want and believe, so that no conflict arises because nobody perceives an interest to defend. The first is loud and attributable; the second is quieter; the third leaves no trace at all, because the subject experiences the belief as his own. A minority that intends to last acts in the third dimension, and descends to the second only where the third has not yet taken effect.

There are three ways to obtain compliance (Etzioni, A Comparative Analysis of Complex Organizations, 1961). Coercion is expensive to sustain and produces alienation: the subject obeys while the person of authority is present and resents him when he is not. Payment is expensive and produces optimisation: the subject obeys as long as the price is right and re-evaluates it constantly. Normative control, obedience because the subject believes he should, costs almost nothing once accepted and produces commitment: the subject obeys unwatched and unpaid, and defends the rule against those who do not.

The first two modes have a second defect that matters more than their cost to anyone who wishes to rule unseen and durably. They are attributable. A prison and a payroll have addresses. A belief installed by association has none; the subject cannot say who put it there, because he does not know that anyone did. The only form of power that is both cheap and untraceable is the one that lives in the subject's own network. Hume identified the arithmetic: force sits with the governed, who are many, so the governors hold nothing but opinion, and on opinion only government is founded ("Of the First Principles of Government", 1741). The resulting arrangement is consent held by culture, schools, press and church, with coercion kept in reserve for the moment consent fails, and the work of maintaining it has a name: the manufacture of consent (Lippmann, Public Opinion, 1922). The coercion reserve is real, but its use is a symptom.

Belief has one property that makes governing it a permanent function rather than a foundation laid once. It decays. Rival narratives appear; a generation that never learned the story replaces one that did; events contradict the story faster than the story can absorb them. And compliance is contagious in both directions: Levi showed that stable rule runs on quasi-voluntary compliance, which holds only as long as each subject believes the others are complying, and collapses in a cascade the moment that belief breaks (Of Rule and Revenue, 1988). A belief system is a flow, not a stock. Whoever depends on it must feed it continuously, against competition, at the speed at which beliefs actually change, which is the speed of repetition.

An objection presents itself. If power is consent, and consent is what violence destroys, how can those whose whole method is the management of belief be the same people who make wars? Arendt stated the objection in its strongest form: power is the capacity to act in concert, it exists only as long as the group consents, and violence appears exactly where power is failing; it can destroy power but never create it (On Violence, 1970).

The answer is that Arendt is right about violence turned inward and wrong about violence turned outward. A ruler who must fire on his own governed has lost them; that is the sign of exhausted power, and a minority that governs by belief never reaches it. Violence against a designated enemy does the opposite. It is the most powerful producer of consent that exists.

Four mechanisms account for this. Conflict produces cohesion: an external enemy tightens the group, silences internal dissent and legitimates whoever leads the fight (Coser, The Functions of Social Conflict, 1956). Its measurable form is the rally effect, the surge in approval of a leader at the outbreak of hostilities. War builds the state: the taxes, conscription, surveillance and administrative reach that populations refuse in peace are granted under threat, and once granted are rarely returned (Tilly, "War Making and State Making as Organized Crime", 1985). War is a narrative event: it supplies heroes, villains and a cause in a single stroke, delivered by the facts themselves, and it moves the range of what a population finds thinkable further in months than persuasion can in years. And war is a means of competition among the powerful: whoever controls the war economy, the reconstruction and the settlement disposes of rivals and captures the returns, an arrangement a general who had led one thought dangerous enough to name in his farewell (Eisenhower, Farewell Address, 1961).

Warfare, finally, is as untraceable as belief. A war is provoked, proxied, financed or allowed to happen; it is almost never declared by the hand that wanted it. Making war is not an exception to the government of belief. It is the government of belief carried out with the largest lever available: the fear, the enemy and the sacred cause, which are the deepest associations every human being carries, activated together.

Because power is a belief held by the governed, because belief is the only control that is at once cheap and untraceable, because belief decays and must be fed, and because organised violence, pointed outward, is the strongest feed there is, a minority committed to lasting has no option but to maintain narratives as a standing, directed and unattributable function. Influence is not one instrument of government among others. It is the substrate the others rest on.

What follows describes the mechanics of that substrate, and then who operates it and how.

Modern influence doctrine descends in a straight line from psychoanalysis. Edward Bernays, who turned wartime propaganda into the profession of public relations (Propaganda, 1928), was Freud's nephew, and his central move was to import the unconscious into commercial and political persuasion: the public does not decide on arguments, it decides on associations it cannot see. His best-known campaign is the one this part uses as its running example. In 1929, on behalf of the American Tobacco Company, Bernays arranged for a group of young society women to light cigarettes in public during the New York Easter parade, and fed the press the phrase "torches of freedom": the cigarette, until then improper for women, was attached to emancipation and adulthood, and the attachment held for half a century.

Three concepts from that tradition map directly onto the mechanics of influence.

The unconscious is the part of the belief system the subject cannot audit. An influence operation can shift a belief system to conviction without the audience noticing that anything has been done to it; that is only possible because most of a belief system consists of associations formed without deliberation. The operator works on the associations the subject does not know he has.

Drives (Freud's Triebe, the French pulsions) are the energy source. Fear, aggression, sexuality, the need to belong, the need to be seen as adult and free: every lever of influence is a way of coupling a message to one of these. A message that engages no drive is a proposition; one that does is a belief. The Freudian mechanisms of defence supply the rest of the vocabulary: identification (adopting the traits of a group or figure) is what group cohesion is built on and what recruitment into any closed order selects for; rationalisation is how a subject resolves the contradiction between what he knows and what he does; displacement, redirecting an affect from its real object to a safer one, is the mechanism behind common-enemy framing.

Jung's collective unconscious is the layer of the belief system that is not individual at all: the part of the associative network that is the same, or nearly the same, in every member of a population, because it was laid down by common human situations: birth, the parent, the enemy, death, the sacred, rather than by any one biography. Its contents are the archetypes and their symbols: the hero, the father, the shadow, the trickster, the forbidden fruit. The collective unconscious is what makes mass influence possible at all. A narrative is written once for a whole audience, not once per subject; that only works if a large part of every subject's network is shared, so that the same words fall on the same associations in each of them, which word-association norms collected from tens of thousands of speakers confirm across a population (De Deyne et al., "The Small World of Words English Word Association Norms", 2019). Archetypes are pre-loaded narrative templates in that shared layer, and a narrative is cheaper to install when it reuses one: the audience already knows how the story ends. The same templates work on an operator's own members as on his audiences; the initiate who has crossed into forbidden knowledge is an archetype too. Mechanically, the collective unconscious is the region of the association space whose geometry is common to the population.

One caveat keeps this section honest. Psychoanalysis is not a predictive science and most of its specific claims have not survived. Its value here is as a vocabulary and a source of templates; what it described in metaphor, cognitive science has since either confirmed in mechanism or replaced.

A belief system is a network of concepts linked by association, each link carrying a weight (how strongly one concept calls up the other) and a valence (whether the pairing is felt as good or bad). Most of the network is not consciously accessible; what the subject experiences is its output: a judgement, a preference, an action, not its structure.

The mechanics are well established. Concepts are stored not as isolated entries but as patterns of connection. A neuron does not link to one other neuron but to thousands, one to many in computer science terms. Each concept is represented by a distributed pattern across many cells, and each pattern overlaps with the patterns of its neighbours. The whole is a weighted graph: nodes are patterns, edges are synaptic connections, and the weight of an edge is the strength of the connection, built up by use.

It helps to read this graph as a dense vector space in which concepts are vectors and proximity is meaning. The distance between two concepts is not a coordinate difference but a path length: the shortest route through the graph from one to the other, where strong connections count as short edges and weak ones as long, the same quantity a shortest-path algorithm such as Dijkstra's computes over a weighted network. Two concepts are close when a strong path joins them, whether directly or through a few intermediate nodes; they are far when every route between them runs through weak links. "Cigarette" sits a short path from "freedom" because the two have been presented together for decades; it sits a short path from "adult" because "freedom" does. Proximity is transitive, and a concept inherits the neighbourhood of its neighbours.

This is what symbolism is, mechanically. A symbol is a concept that sits at short path length from a large cluster of others, a hub in the graph, so that activating it activates the whole cluster at once: fire, and with it warmth, danger, home, destruction, the sacred. Symbols are powerful for the operator for two reasons. They reach many concepts through one word or image, and their neighbourhoods are largely shared across individuals, because the associations that made them hubs were laid down by common experience rather than private biography. That shared geometry is the collective unconscious, read as structure: the region of every subject's space in which the distances are the same. The picture is an analogy, the brain does not store coordinates or run Dijkstra, but it is a faithful one, because the associative network behaves like a similarity space with exactly this metric, and the cortex itself has been mapped as one (Huth et al., "Natural Speech Reveals the Semantic Maps That Tile Human Cerebral Cortex", 2016).

The dynamics follow from the geometry. Activating one concept spreads activation to those near it, decaying with path length (spreading activation, Collins and Loftus, "A Spreading-Activation Theory of Semantic Processing", 1975); concepts that are repeatedly activated together become more strongly linked, which shortens the path between them (Hebbian learning, Hebb, The Organization of Behavior, 1949: what fires together wires together); and a concept repeatedly paired with a pleasant or unpleasant experience acquires that valence itself (evaluative conditioning; De Houwer et al., "Associative Learning of Likes and Dislikes", 2001). Larger structures like thinking schemas or stereotypes are simply dense regions of the graph, clusters of short paths that activate as a block.

Three consequences follow.

First, a belief is moved by moving its neighbours. Arguing against the proposition "smoking is harmful" is expensive and usually fails; adding "adult", "free" and "attractive" to the region around "cigarette" is cheap and succeeds. This is why programs, described in Part V, are written as belief shifts rather than as arguments, and why what a carrier is given to hold is a set of associations, not a thesis to defend.

Second, associations are additive. New links do not delete old ones; they sit beside them. A health warning does not remove "cool" from the cigarette's neighbourhood, it adds "danger", and the network then has to reconcile the two.

Third, the network is physical and therefore inert. It changes at the speed of repetition, not at the speed of information. This is the biological basis of persistence: a belief that has been installed by repetition survives the end of the program that installed it, because the links are now structural.

Belief systems do not only store associations; under certain conditions they reinforce themselves. The cigarette is the cleanest documented case, because its belief system was installed by one campaign, contradicted by public information, and then rebuilt by the subjects themselves into something stronger than before.

Before awareness of the health risks. The cigarette's neighbourhood was built by decades of advertising and cinema: cool, social; freedom and adulthood; relaxation, a hobby, something natural. A plant, grilled, close to fire and to the outdoors. Nothing in that region was aversive. Smoking was a belief system held at conviction across most of the adult population.

After awareness of the health risks. From the 1960s (the 1964 US Surgeon General's report), and aggressively from the 2000s with graphic warnings on packs, a second set of associations was added: cardiovascular disease, cancer, damage to others. In aggregate, the warnings worked. Smoking declined. But for the subject who continued, the network now contained two incompatible regions, and it forced a resolution. The old associations were not deleted; they were joined by new ones that fit both: transgression, rebellion, risk-taking, and above all a claim of exception: it happens to others, I am above it, which is an evolutionary-fitness signal, the display of surviving what harms the ordinary. The Hollywood associations remained intact underneath.

Why does the network resolve the contradiction this way rather than by quitting? Because "cool", "free", "adult" and "rebel" are already close to one another in the subject's space. Closer, for most people, than "cool" is to "prudent". The danger association, arriving into that region, lands among concepts that convert it: danger next to freedom reads as courage. This is cognitive dissonance in Festinger's sense (A Theory of Cognitive Dissonance, 1957) resolved at the cheapest point, and it is mostly unconscious; the subject does not decide to become a rebel, he finds that he is one.

The loop. The resolution has a second effect. Once the subject has adopted the associations that make continued smoking coherent, he has adopted an identity. He self-categorises as a smoker; Turner's self-categorisation (Turner et al., Rediscovering the Social Group, 1987), a person taking on the attributes of a group as his own; and with the category come its whole behavioural framework and its worldview. The fatalism, the short horizon, the pleasure taken against advice. Addiction then holds the loop closed: the drive to smoke keeps the identity active, the identity keeps the drive legitimate. The brain traps itself, from a chemical dependency into a belief system, and from the belief system into a stereotyped set of behaviours. There is strong evolutionary pressure behind this: coherence between what one does and who one is was, for most of human history, a survival trait, and the belief network defends it at the cost of the facts.

The addicted smoker is, in this sense, subjected to smoking and governed by his drives. He is what this essay will call a nihilistic idealist: he holds, with real conviction, a belief system, freedom, authenticity, contempt for prudence, whose actual content is the negation of his own interest.

Why this matters to the operator: a self-reinforcing belief system is the most valuable object a program can produce, for three reasons.

It classifies. Smokers are a segment with a known worldview, known drives, known reward pathways; the tobacco industry's internal research, released in litigation under the 1998 Master Settlement Agreement, segmented them by psychological profile decades before "psychographics" was a marketing term. A subject who has self-categorised has done the operator's target-audience analysis for him.

It predicts. The belief system entails behaviours beyond smoking: purchases, risk preferences, receptivity to certain narratives and hostility to others, so the segment is exploitable by programs that have nothing to do with tobacco. A belief system installed by one program is a channel for the next; profiles derived from one behaviour measurably raise persuasion on unrelated ones (Matz et al., "Psychological Targeting as an Effective Approach to Digital Mass Persuasion", 2017).

It sustains itself. Once the loop is closed, the program that created it can end; the subjects maintain the belief system for it, and recruit into it. The belief persists, and part of the segment has become its vectors. The recognition of the cigarette's harms, arriving as it did into an established belief system, produced a smaller but far more committed segment; higher yield per subject, for the operator, than the diffuse mass it replaced.

The general form: any belief system with an identity-bearing or addictive behaviour at its centre is self-reinforcing, and contradiction from outside rarely weakens it, and often leaves it intact, because the network absorbs the contradiction as a new reason for the identity (corrections work least on identity-linked beliefs: Nyhan and Reifler, "When Corrections Fail", 2010; on the rarity of outright backfire, Wood and Porter, "The Elusive Backfire Effect", 2019). Building such loops is the highest objective a psycho-behavioural program can be given.

The cognitive-science literature lists several dozen biases and effects. The framework uses at its core a subset, chosen because each corresponds to a stage of the ladder by which a belief is installed: awareness, acceptance, conviction, advocacy. They are grouped here by what they serve.

Awareness: getting the association in.

Illusory truth effect and processing fluency (Hasher et al., "Frequency and the Conference of Referential Validity", 1977). A claim heard several times feels truer than one heard once, and anything easy to read, hear or picture is judged more true and more likeable. Repetition substitutes for evidence; simplicity substitutes for argument. This is why carriers are made to hold one narrative in fixed words over a period: the words must recur unchanged.

Mere-exposure effect (Zajonc, "Attitudinal Effects of Mere Exposure", 1968). Familiarity alone breeds liking. A name, symbol or narrative that has merely been seen often is preferred to one that has not. Program reach is measured for this reason: exposure has an effect before any content is processed.

Acceptance: making the association credible.

Authority bias and the halo effect (Milgram, "Behavioral Study of Obedience", 1963; Nisbett and Wilson, "The Halo Effect", 1977). Claims from credentialed or official-looking sources are accepted with less scrutiny, and competence in one domain spreads to credibility in another. This is why an operator's carriers are chosen among those whose public position already carries the halo, or catapulted to top positions and given exposure for the same reason.

Ingroup favouritism (Tajfel, "Experiments in Intergroup Discrimination", 1970). A message from "one of us" is trusted; the same content from an outgroup is rejected. Hence the matching of carrier to audience by position, and the value of holding voices across the entire political spectrum, dissidents included: every segment must have a carrier it recognises as its own.

Social proof, pluralistic ignorance and the spiral of silence (Asch, "Effects of Group Pressure upon the Modification and Distortion of Judgments", 1951; Noelle-Neumann, "The Spiral of Silence", 1974). If many appear to believe something it is taken as likely true; people who privately disagree assume everyone else agrees and stay silent, which makes the apparent majority larger. The appearance of an accepted idea is enough, it need not be real.

Conviction: attaching the association to identity.

Confirmation bias and identity-protective cognition (Kahan, "Ideology, Motivated Reasoning, and Cognitive Reflection", 2013). People seek and accept what fits existing beliefs, and beliefs tied to group identity are defended against evidence; facts are evaluated by who benefits rather than by whether they are true. This is where the loop begins: once a claim is part of who the subject is, contradiction is an attack on him.

Cognitive dissonance (Festinger, A Theory of Cognitive Dissonance, 1957). Discomfort from holding conflicting beliefs is resolved by changing whichever belief costs least. The cigarette is the worked example: withdrawal side effects make quitting too expensive; it is more comfortable to extract gratification from the risk itself.

Selective exposure and the individual echo chamber (Nguyen, "Echo Chambers and Epistemic Bubbles", 2020). Selective-exposure research shows that people choose sources that agree with them (Hart et al., "Feeling Validated Versus Being Correct", 2009) and avoid those that do not, and the sources they keep are in turn selected by the same preference; over time a subject's network receives almost only inputs that shorten the paths it already has. In the geometry of the belief system this is a subject feeding his own neighbourhood: every new association arrives pre-aligned, and the region hardens. Two types of situations should be distinguished, and Nguyen's distinction between the epistemic bubble and the echo chamber names them. In the first type the subject simply never meets the contrary view. In the second one, he is given a contrary view and has been given a reason to distrust it in advance, so that contact with the outside no longer breaks the enclosure but confirms it. That is the echo chamber. Programs aim at the second, because it is the form that survives exposure.

Continued influence effect (Lewandowsky et al., "Misinformation and Its Correction", 2012). A correction rarely erases the original claim; people keep reasoning with what was retracted. This is the biological basis of persistence, and the reason an operation is stopped by attribution exposure rather than by refutation: refutation does not undo the work.

Advocacy: turning the subject into a vector.

Affect heuristic, fear appeals and moral contagion (Slovic et al., "The Affect Heuristic", 2002; Brady et al., "Emotion Shapes the Diffusion of Moralized Content in Social Networks", 2017). Feelings about a topic substitute for rationality; reminders of threat increase attachment to the ingroup and to strong leaders; morally charged content spreads farther and is remembered better than neutral content. Advocacy-level objectives are reached through emotion, not argument, because emotion is what makes a subject transmit.

Commitment and consistency (Freedman and Fraser, "Compliance Without Pressure: The Foot-in-the-Door Technique", 1966). Small agreements lead to larger ones; a belief stated in public is held more rigidly and spreads by proximity. Once someone has acted on a belief, abandoning it costs face. Turning a sub-segment into vectors is a commitment device: the act of spreading the belief locks it in the spreader and the audience.

The second family of levers comes from political science, communication theory and influence doctrine rather than cognitive science. They sit on top of the individual mechanisms, but they operate across programs and over decades rather than on a single message.

Shifting what is thinkable. The Overton window is the range of ideas the public considers acceptable at a given time; the term is a practitioner's, and its peer-reviewed form is Hallin's division of public discourse into spheres of consensus, legitimate controversy and deviance (The Uncensored War, 1986). It is moved not by arguing for the target position but by pushing a more extreme one, which makes the target look moderate; normalisation does the same by repetition, each exposure to what was shocking making it ordinary, and salami tactics by steps too small to resist. The first rung is the hardest and has its own literature: turning the unthinkable into the merely radical is a Tabubruch, a shifting of the boundaries of the sayable (Bourdieu, Outline of a Theory of Practice, 1972, on turning unspoken doxa into contestable heterodoxy). It requires a norm entrepreneur willing to make the first move, and it works because most people conceal what they think until someone says it aloud, at which point the concealment collapses in a cascade (Kuran, Private Truths, Public Lies, 1995). Where no word exists for the idea, coining the word is the whole act: a thought with no frame is hard to hold and harder to pass on (hypocognition, Levy, Tahitians: Mind and Experience in the Society Islands, 1973). No single program moves the window; a portfolio of programs scheduled over years does, and the radical flank (Haines, "Black Radicalization and the Funding of Civil Rights", 1984), an extreme faction that makes the moderate demand acceptable, is a program run for the benefit of another.

Controlling what gets discussed, debated and thought about. Agenda-setting (McCombs and Shaw, "The Agenda-Setting Function of Mass Media", 1972) is the finding that media do not tell people what to think but what to think about; gatekeeping is control over which facts and voices reach the audience at all; cultivation (Gerbner, "Toward Cultural Indicators", 1969) is the slow shaping of a population's sense of what is normal, crime, threat, wealth, by long-term exposure. These operate on media rather than on messages: an operator's media and platform assets are not primarily there to say things but to spread and properly target what is discussed by sustained means.

Narrative and meaning. Framing (Entman, "Framing: Toward Clarification of a Fractured Paradigm", 1993) chooses the vocabulary that pre-decides the judgement ("relief", "collateral"); a master narrative supplies a simple story with heroes, villains and a cause, into which events are then slotted; loaded language offers virtue words no one can oppose; the thought-terminating cliché (Lifton, Thought Reform and the Psychology of Totalism, 1961) is a stock phrase that closes discussion. A single master narrative, held by every carrier, is what turns scattered events into a coherent picture in the audience's network.

Managing truth and doubt. Manufactured doubt presents a settled question as contested, lowering confidence without winning the argument; the tobacco industry's own method ("doubt is our product", Brown and Williamson memo, 1969), and the counter-move to public awareness. The firehose of falsehood (Paul and Matthews, "The Russian Firehose of Falsehood Propaganda Model", RAND, 2016) floods the audience with high-volume, contradictory messaging so that verification is overwhelmed and the outcome is resignation rather than belief; apathy induction is the recognition that disengagement is often the actual goal. Saturation is used when persuasion has failed or is unnecessary. Plausible deniability and the limited hangout exploit this mechanism to manufacture disengagement; they are deniability applied to operations. Reflexive control (Vladimir Lefebvre, Conflicting Structures, 1967; Thomas, "Russia's Reflexive Control Theory and the Military", 2004), feeding the audience the premises from which it will reach the desired decision by itself, is the most refined form.

Exploiting opposition. Manufactured opposition runs the resistance so that it stays ideologically harmless; wedge issues split an audience along a divisive topic; common-enemy framing (Coser, The Functions of Social Conflict, 1956) produces unity through an external threat, real or inflated. They model seemingly opposite voices in one debate, which is why an operator keeps "dissidents" and "extremists" among his own voices.

Sealing the audience. At the level of the mass, the echo chamber is a structure rather than a habit. Homophily (McPherson et al., "Birds of a Feather: Homophily in Social Networks", 2001), the tendency of people to associate with those like them, sorts a population into clusters that mostly talk to themselves; platform recommendation systems, which optimise for engagement, amplify the sorting at scale. By how much exactly is disputed: the largest field experiments found that people's own choices do most of the work and that removing the algorithm does not undo it (Guess et al., "How Do Social Media Feed Algorithms Affect Attitudes and Behavior in an Election Campaign?", 2023). The resulting filter bubble delivers to each cluster a different picture of the world. The chamber proper adds distrust: a cluster whose members have been taught that outside sources are hostile, so that the cluster polices its own boundary. For an operator this is the most economical object there is. A sealed segment is a channel that needs no gatekeeping, a container that maintains itself, and the social form of the smoker's loop: that belief system was individual, the echo chamber is the same loop run by a group, each member reinforcing the others' associations and supplying the social proof that keeps the enclosure credible. It is also the mechanism of polarisation (Sunstein, "The Law of Group Polarization", 2002). Two chambers fed opposite narratives drift apart until no common ground remains, which is the wedge issue carried to completion.

Hidden in plain sight. Finally, the mechanism by which the whole apparatus stays unrecognised: secrecy and deniability seen from the outside. Information passes through four stages before it acts: accessible, noticed, understood, believed. An operator works on all four. Data can be fully public and still hidden if no one looks, if no one has the concept to read it as meaningful (Fricker, Epistemic Injustice, 2007), or if the messenger has been discredited in advance so that public availability lowers credibility rather than raising it (Cohen, States of Denial, 2001, on the three forms the rejection takes: it did not happen, it is not what you say, it does not matter). It is the Cassandra position: true, public and disbelieved. What everyone knows but no one has said in public is a hidden transcript (Scott, Domination and the Arts of Resistance, 1990) or a public secret, knowing what not to know; it becomes actionable only when it becomes common knowledge, known by all to be known by all (Chwe, Rational Ritual, 2001), which is precisely the transition the apparatus prevents. Control of interpretation is cheaper and more robust than control of access, although both are useful. This is why the apparatus need not restrict access to most of its methods: what it controls is the key to reading them. A truth already circulating and discredited, mixed with false facts and unbelievable rumours is better protected than one under seal and clearly explained.

Now that we have a comprehensive overview of the behavioural, cognitive and psychological levers, we must map the organisation that applies those concepts and sustains their application over time.

A circle, in the sense used in this essay, is a closed group of people united in a common design, promoting their shared views or interests in an ideology, a state, or another community, with discretion and usually without the knowledge of those outside the group.

The people this part describes are defined by position and by mastery, not by wealth alone. They are of two kinds: those who head institutions, states, large companies, clergies and the leading families; and those who master the domains these institutions run on, the leading scientists, doctors, intellectuals and financiers: leaders and specialists. Together they form a small group holding a vastly disproportionate share of the decisional power a society runs on.

What follows is a model grounded in the literature and in testimonies: it describes the form the documented cases point to rather than any one of them. To understand the scale of the practices described in this essay, we must first describe the framework inside which they take place. Sociology has long documented that they group themselves into clubs and closed societies: the Ivy League secret societies, Skull and Bones being the best known, are the usual example; they serve as antechambers to a more established national body where the real ruling group, made of decision makers and the oligarchy, gathers (Rothkopf, Superclass, 2008).

Such a body is what we call a circle, after the "higher circles" of Mills (The Power Elite, 1956).

It is equally documented that they appoint one another to corporate and institutional leadership positions. The most visible form is dynastic succession: wealthy families and the aristocracy placing their children at the head of their corporate empires, of which the Arnault succession at LVMH is a public example.

Circles are organised by country, yet share interests internationally, in particular in matters of governance.

Modern circles have one tier and one office: a flat body of members with equal status and voting rights, who decide in session, and one member who is the circle's negotiator, usually the head of state but not always. He carries the members' mandate out and the terms back, and the decision stays with the members. He holds imperium from his office and no auctoritas of his own. The Doge of Venice was elected by the patriciate, bound at each election by a written promissione the body revised, and represented the Republic without governing it (Lane, Venice: A Maritime Republic, 1973). The Grand Pensionary of Holland carried the mandate of the States into every negotiation and brought the terms back for their decision (Rowen, John de Witt, Grand Pensionary of Holland, 1625–1672, 1978). The members author the primary artifact of the sheet, which Part V calls the doctrine, and what is written under it answers to them. Inside the circle speech is free and knowledge whole: a member holds everything the circle holds (Bagehot, The English Constitution, 1867). What the circle holds whole is its own knowledge, principles and artifacts, held in persons, and with them everything its instruments return, read by every member through the artifacts regardless of who is assigned to them.

The membership is made of the top individuals in their respective domains and of the international oligarchy. This includes:

These categories are not chosen for prestige but for interlock: each is a seat on an institution the circle needs to act through, and a member counts by the seats he holds, not by his wealth (Useem, The Inner Circle, 1984; Mizruchi, "What Do Interlocks Do?", 1996).

The circle's leadership mirrors the leadership of top institutions: heads of clergy, heads of state, chief executives of large companies; and the heads of the leading oligarchic families. In public, a member's conduct follows the logic of the Instructions defined in Part V: he holds the narrative core and acts within the bounds set for his position, the choice of action and timing being left to his judgement. This is especially true of politicians and the high clergy, whose rhetorical and representational strategies are described in later parts.

This structure is not hypothetical. In 1981, during a financial investigation, Italian police found the membership list of Propaganda Due, a body Licio Gelli ran under the cover of a private lodge: over nine hundred names, among them the heads of both intelligence services, some fifty generals and admirals, sitting ministers, dozens of parliamentarians, the leading bankers, the owner and the editor of the largest newspaper, and a future prime minister. A second search produced its written programme, the Piano di rinascita democratica: control of the press and the parties through selected men, reform of the judiciary, a rewriting of the constitution, with budgets attached. The masonic label was cover, not structure; P2 had one master, a chain of instruction and a membership chosen for the offices it held. Read against this essay, the list is a membership built for interlock and the plan is a doctrine sheet defined in Part V (Commissione parlamentare d'inchiesta sulla Loggia P2, Anselmi report, 1984). It is the closest documented instance of a circle, and it shows where the form fails: it was national, it kept a list, and it was found.

The form is old. In Roman law a body of persons, the universitas, persists while its members change. In a corporation, Ulpian writes, it does not matter whether all remain, or all have been replaced, and from that body descends the one that never dies (Kantorowicz, The King's Two Bodies, 1957). The priestly colleges of the Republic gave the body its working form: pontiffs, augurs and the keepers of the Sibylline books, co-opted by the college itself, holding office for life so that the religious and the political leadership interlocked in the same persons, each college keeping the books of a discipline that was its own and opened only under rule (Beard and North, Pagan Priests, 1990). The circle is that form: a body with principles, a procedure and a memory, the last being the part Hobbes gave his artificial man in its counsellors (Leviathan, 1651, Introduction), with the deity moved within.

Circles are governed by shared core principles, split between primary and secondary. The primary principles are those of existence: a principle is primary when its violation ends the circle. The secondary principles are those of action: they derive from the primary ones, the circle survives their violation, and they are ranked, Mastery first, since no other can be exercised without it.

Primary principles

Secondary principles

Alongside this training, members are inducted into an internal belief system, that of the "self-divine": not a conviction professed but a sense lived, a type of religion with the deity moved within, so that members experience themselves as of a different nature from those they govern, consecrated rather than merely successful. Roman divinity was of this kind, a relative standing on one continuum of honour rather than a claim about nature (Gradel, Emperor Worship and Roman Religion, 2002). It is built from three sources. Ritual, since a group that assembles under fixed ceremonies experiences its own cohesion as something sacred and attributes it to itself (Durkheim, The Elementary Forms of Religious Life, 1912). Secrecy, since possessing what others are excluded from produces a consciousness of exception independent of what is possessed (Simmel, "The Sociology of Secrecy and of Secret Societies", 1906). And impunity, since power reliably lowers empathy and raises the sense of control and entitlement, so that what is not punished comes to feel deserved (Keltner et al., "Power, Approach, and Inhibition", 2003). The result is the doctrine every ruling group needs, that its power is deserved and its fortune just (Weber's theodicy of good fortune, Economy and Society). This belief system is the cohesion principle made personal: it is what binds members to one another more strongly than to any outside loyalty.

Sheet

The circle's constitution is its principles, the procedure that applies them, and the artifacts that hold what it knows and acts by. Together they are the national object called the circle's sheet:

Sheet(circle) = { principles, procedure, artifacts }

No field grants the power to enforce: enforcement is the members applying the procedure in session, their standing named by the Authority principle and their arm the artifacts they collectively own; and the arrangement holds because members enforce it on one another, a circle being the union of principles of conduct and procedures for applying them, with no further rule behind (Hart, The Concept of Law, 1961), self-enforcing where no one stands above its members.

The artifacts are the circle's memory and tools. Everything the circle keeps is an artifact: a text with one or many assignees, named under Assignment below, and actionable by design, since an artifact exists to be acted on and inherits the actions taken under it. The procedure defines three base artifacts:

Artifacts(procedure) = { sheet, primary artifact, calendar }

The primary artifact is the one artifact written from the principles directly. It states the goals they require of the body, each with a term, and everything else the circle produces is written under it. Part V gives it its form and calls it the doctrine. It is defined under its own procedure below. The calendar is the artifact that holds time: the sequence of procedures taken and the sequence of procedures planned, the terms of seats, the sessions of assemblies, the cycle on which the primary artifact is revisited, the dates and schedules of the artifacts it names, and the clocks of everything that runs under them. The occasion of every procedure is read from it, and every procedure taken is written to it. The pontiffs kept such a calendar for Rome, the fasti, the days on which public acts could be done and the magistrates who had held office year by year, held by the college and opened to the city only late (Rüpke, The Roman Calendar from Numa to Constantine, 2011). The monastic rule did the same with its hours, and the schedule has been since then the device by which a body makes its own conduct predictable to itself (Zerubavel, Hidden Rhythms: Schedules and Calendars in Social Life, 1981). Everything else the circle produces is a secondary artifact, proper to each circle and defined in Part V.

The artifacts are not inert. Each holds a state and a clock, and acts on others. The primary artifact bounds the artifacts written under it. An artifact produces others and re-executes on them when it is revised. What an artifact captures while it runs re-enters the models of the one above it. What holds across them is distilled upward into the next primary artifact. The calendar times all of it. The artifacts are therefore a graph. The nodes are the artifacts. An edge is the output of one entering the state of another.

Artifacts(circle) = { nodes: artifacts, edges: output of one entering the state of another }

The graph is open. A circle adds a kind of artifact by adding a node and its edges. The procedure does not change, because it names assignees and occasions and never kinds. Artifacts are living objects in this sense: objects that do something to the next objects. Institutional analysis knows the structure as a network of adjacent action situations, where the outcome of one is an input of the next (McGinnis, "Networks of Adjacent Action Situations in Polycentric Governance", Policy Studies Journal, 2011). Engineering knows it as a feedback structure, stocks joined by the information each sends the others (Forrester, Industrial Dynamics, 1961).

Together the artifacts are the memory a lasting body needs beyond its persons, held in texts that outlast them (Jan Assmann, Das kulturelle Gedächtnis, 1992). In machine terms the artifacts are running processes rather than files: each holds its state, acts when its clock says, and writes into the others. The calendar is the clock, the members the cores, each with the whole in cache. Inside the circle nothing is closed: every member reads every artifact, its structure as much as its text and everything its resources return, and members see the world through them. There is no asymmetry among members. The interface is perfect: it gives every member everything, regardless of assignment.

The circle has no rules of conduct. Conduct is weighed by the principles, and a principle has weight where a rule applies or does not (Dworkin, "The Model of Rules", University of Chicago Law Review, 1967). What the circle writes down is how the body itself acts. A procedure defines one admissible action: who may take it, on what occasion, and by what decision it passes. The principles weigh actions; the procedures say which actions exist and whose they are. A body of peers is bound by its forms and not by commands, which is what separates collegial government from bureaucratic rule (Weber, Economy and Society, 1922). Each procedure is a statement of one form:

Procedure = ( who, may | must | must not, action, occasion and decision, or else )

The form is the grammar of institutional statements: attributes, deontic, aim, conditions, and the or-else that makes a statement a rule rather than a norm (Crawford and Ostrom, "A Grammar of Institutions", American Political Science Review, 1995). The circle has one or-else: an act outside its form is null. Each procedure below is given in words and then in this form. Its occasion is read from the calendar, and its taking is written to it.

States. A person stands in one of three states toward a circle: candidate; member; negotiator, the member holding the office. There is no state after member: knowledge once held cannot be returned, so a member is a member for life.

States = { candidate, member, negotiator }   the who of every procedure below

Admission. Candidate requires a sponsor who is a member; member requires selection and the vows sworn on the principles and procedures. The members have the final word on admission, in session, by majority of all members.

Admission = ( members, may, admit a candidate as member, on a member's sponsorship, selectionand the vows, in session, by majority of all members; else null )

Vote. Members decide in session the matters the procedure assigns them, by majority of all members, and elect the negotiator. No vote touches a principle: a principle is held or failed, never amended, and a circle that fails a primary one ends. The procedure is applied by the members in session and moves by its practice; it is rewritten only at founding.

Vote = ( members, may, decide what the procedure assigns them and elect the negotiator,in session, by majority of all members;anyone, must not, amend a principle; else null )

Session. The members meet in session on the dates written in the calendar. Outside those dates a session is convened only on a principle being failed, on the call of members, and for that matter alone. Whoever decides the exception is sovereign over the body (Schmitt, Political Theology, 1922), which is why no one here decides it: the occasion is a principle failed, and anyone may read it. Members vote only in session. The session paper and the applicative plans, described in the next section, are distributed in session, each assigned under Assignment. The session's acts and votes are written to the calendar. A rule and a schedule of sittings are what make an assembly a body rather than a crowd (Bentham, An Essay on Political Tactics, 1791).

Session = ( members, must, meet in session on the dates in the calendar;members, may, convene a session on a principle being failed, for that matteralone, and vote only in session, the acts and votes written to the calendar;else null )

Force. An artifact is in force or it is not. It enters force when its author writes it in its form, naming what it binds and whom it assigns, and the assignee signs it. It leaves force at its sunset, where its form names one, or when its author revises it. An artifact whose form names no sunset stands until revised. Its author may revise it on the occasion its form names, and on no other; a body that can revise at will can promise nothing (Kydland and Prescott, "Rules Rather than Discretion: The Inconsistency of Optimal Plans", Journal of Political Economy, 1977). What a revision does to what was produced under the artifact is set by the artifact's own form. What an artifact produces is an artifact, assigned under Assignment below and named in Part V. Outside force an artifact is null and archived: a text, assigned to no one, binding no one.

Force = ( author, may, put an artifact in force by writing it in its form, naming what it bindsand its assignees, the assignee signing it, and revise it on the occasion its formnames and on no other, until its sunset where it has one; else null and archived )

Primary artifact. The members author the primary artifact in session, with the specialists they convene, deriving its goals from the principles, and it stands: it names no sunset, it is revisited on the cycle written in it and on nothing else, and the cycle is very long. Its length is what makes the body durable (Williamson, "The New Institutional Economics: Taking Stock, Looking Ahead", Journal of Economic Literature, 2000). A revision binds forward: what was opened under it runs on under it as it stood at opening. For each goal, the primary artifact names the artifact that serves it, the date by which it must be in force, its schedule, the emergency on which the members may revise it before its schedule, what it produces and what its revision does to that, and what runs if it is not in force by its date. The dates are written to the calendar when the primary artifact enters force. One procedure that says who makes the primary artifact and how is enough to put every artifact below it in motion (Hart, The Concept of Law, 1961). An artifact that names no date for what must be written under it is not carried out (Gersen and O'Connell, "Deadlines in Administrative Law", University of Pennsylvania Law Review, 2008).

Primary = ( members, may, author or revise the primary artifact, on the cycle written in it,in session, by majority of all members, the revision binding forward;members, must, put each artifact the primary artifact names in force by its dateand on its terms; else the default it names runs )

Succession. The negotiator is elected by the members for a term. The office falls vacant by term, by death, or when he leaves it, and is refilled by election, the elected being formed beforehand under the Succession principle. A member who leaves the office stays in the circle.

Succession = ( members, must, refill the negotiator's office by election, on term, death ordeparture, the elected formed beforehand; else Continuity fails )

Assignment. Every artifact is the circle's, no member owns one. What varies is who acts on it. An assignment is an artifact that binds one or many members to act on another. It names the task, the actions and resources of that artifact they may order and allocate, the bounds inside which they decide, its clock, and the conditions on which it passes to others. An artifact holds resources as well as actions, and the resources are the members' own: a seat, an office, an asset, a holding, time, brought under the artifact by assignment. The author of an artifact writes its first assignment at creation and names the assignees. An assignee takes it by signing. The members in session author the primary artifact and are its assignees as a body. A secondary artifact is authored on the terms of the artifact above it and assigned to the members who sign it. The sheet and the calendar are the circle's, written to by whoever applies a procedure. Assignment distributes actions, never knowledge: every member reads every artifact and every assignment, regardless of who holds it. The assignees answer for the task to the members in session for as long as the assignment is in force, for what they ordered and never for the hand that executed it (Rome Statute, 1998, article 28). The form is the oldest on record. A body that owns everything in common entrusts its goods to brothers who answer for them, keeps the list, and records the handover when one succeeds another (Rule of Saint Benedict, chapters 32 and 33). The corporation of Roman law owned as a body and acted through the agent it appointed (Digest, 3.4.1.1 and 3.4.7.1). What is assigned is the task and its bounds; what is left to the assignee is the choice within them (Simon, Administrative Behavior, 1947).

Assignment = ( author, must, write an artifact's first assignment at its creation, naming the task,the bounds, the clock and the conditions on which it passes;assignees, must, sign it, order the actions and allocate the resources the tasknames, and answer for it in session while it is in force;assignees, may, pass the assignment on the conditions it names, and on no other;anyone else, must not, order or allocate what an artifact holds;every member, may, read every artifact and every assignment; else null )

Assent. The graph of artifacts is open beyond the circle. The members in session may tie an artifact of the circle to a resource outside it, held also by an artifact of another circle, through which the two artifacts act together. Each circle keeps its own artifact. What passes between the circles is what the resource's own form fixes and nothing else. The resource is read by both bodies and used by each in its own way, which is what lets separate bodies work on one object (Star and Griesemer, "Institutional Ecology, 'Translations' and Boundary Objects: Amateurs and Professionals in Berkeley's Museum of Vertebrate Zoology, 1907-39", Social Studies of Science, 1989). At this level there are artifacts and what they hold, and nothing else: the procedure is an institution in the strict sense, the rules of the game, and the bodies of the world are players, which enter the circle only as resources its artifacts hold.

Assent = ( members, may, tie an artifact to a resource outside the circle, held also by anartifact of another circle, through which the two artifacts act together,in session, each circle keeping its own artifact; else null )

These principles, procedures and artifacts are the constitution of the circle. From them, the members in session craft the corpus of doctrines members enact in office, with experts in psychology, engineering, intelligence and management. The experts work in compartmented units, described in Part VIII: each holds a task and sees neither the design nor the whole. They write against a view of the constituency that is complete and live, held by the base of the clandestine instruments described in Part VIII. That corpus is specified as a doctrine sheet and transformed using playbooks described in Part V.

Structures that resemble the circle exist, and their members intersect with the circle's: lodges, orders, clubs, boards, services. They are not circles. Each defines an asymmetry of information among its members, by degrees, by need to know, or by a committee that knows more, and each defines ownership of its artifacts by some members against others. Both contradict the definition above. A circle has one interface and no asymmetry among members, and its artifacts are the body's. A member may belong to several such bodies; the circle is the one in which he holds everything. The distinction is old. A secret society is defined by the degrees of initiation that grade what its members know (Simmel, "The Sociology of Secrecy and of Secret Societies", American Journal of Sociology, 1906), and the committee that knows more is the fate of every body large enough to need one (Michels, Political Parties, 1911). The circle is what remains when neither applies: a body small enough to need no degrees and no committee. Where a resembling structure holds asymmetry or ownership, it is a different body, and the circle meets such bodies as it meets any others, through the resources its artifacts hold, under Assent. Their members are reached differently. A circle member who sits in a lodge or on a board carries derived artifacts into it, and those around him can be instructed under it, as the deputies and staff of members are. They act within the bounds an artifact of the circle sets and receive the instructions written under it, as Part V describes. What reaches them is written under an artifact the circle keeps; the artifact itself never leaves the circle.

An assembly is a session under the Session procedure of the sheet: its date is read from the calendar and its votes are written to it. The circle's oligarchs own large country estates and host regular assemblies in their castles. Less formal gatherings take place in private mansions and on islands built like temples.

Fixed ceremonies open and close each session (compare Domhoff's field study, The Bohemian Grove, 1974; on ritual and solidarity, Collins, Interaction Ritual Chains, 2004). The substantive part is a dinner followed by the study of a session paper, shared by all, and of a per-member applicative plan, both authored by specialised members under the primary artifact. The session paper carries the matter of the session, described thoroughly in Part V; the applicative plan is the member's guidelines bound to his position. Seating is mixed: politicians, scientists, oligarchs and clergy share tables regardless of seniority, domain and wealth, reshuffled at each assembly, while speaking order follows seniority: the session's designated specialists present the topic, then it is debated.

The building blocks of an assembly are as follows. Ceremonies open with a speech from a leader presenting the session; a concert or a show may also take place. Papers are distributed while the meal is being prepared, and preliminary discussions begin. Dinner follows, with discussion and debate. After any voting sessions, secondary discussions take place, along with consultations between members on the assembly's topic and on their applicative plans. The last stage then opens: members join salons and private rooms, and parties can take place according to preferences.

Less formal gatherings also take place between subgroups of the circle, by affinity, to refine guidelines and teachings and for recreational or business purposes, side clubs also exist.

In application of the succession principle, circles run recruitment sessions in top universities such as Harvard, and in the leading European and Asian institutions. Recruits from outside established circle families are selected cautiously on cognitive ability, personality and beliefs, in line with the principles above (sponsored rather than contest mobility: Turner, "Sponsored and Contest Mobility and the School System", 1960; Bourdieu, The State Nobility, 1989). They are assigned mentors who provide warm introductions. In this way circles replenish their ranks and propel new recruits to top positions in existing and new organisations. A striking pattern is that of young billionaire founders: introduced to the circle through their family network, their ventures are subsequently heavily financed and sustained at market leadership. These strategies are described more fully in later parts.

A member's primary training is standardised; a secondary training is then tailored to his specialty. The form is that of every body that forms an elite, a common stage and a specialised one: the grandes écoles and the grands corps, the boarding schools and the boards, the graded degrees of the secret societies, the basic course and the branch school (Bourdieu, The State Nobility, 1989; Cookson and Persell, Preparing for Power, 1985; Simmel, "The Sociology of Secrecy and of Secret Societies", 1906). What the record does not show is the syllabus, and the framework fills it: in application of the mastery principle, senior members act collectively as tutors to newcomers, and the mentors of the recruitment stage transmit circle-specific learning material, including the classified knowledge and methods described in this essay. Two stages follow from the principles: a standardised first stage is also the probation on which admission is decided, and the methods described here are given only after it, so that no recruit holds what the circle knows before the circle knows him.

The constituency is not held in one model but in several, one per domain. Each domain is a vectorial space of its own, an instance defined by its axes: a behavioural space, an economic space, an institutional space, an informational space, and a model is that space typed and stored, holding the whole constituency within it. In what follows, we shall keep the behavioural model as the running example.

a. The space

Formally, a model is the stored algebraic representation of one domain: the space itself, its axes and its metric. It is a dense vector space in which every element is a point with coordinates, stored alongside the transformations that act on it. The space is a function of time. Axes appear and die with the issues they measure, the metric moves with the price of attention, and the points drift between measurements: a dynamical system. The behavioural space is no invention: it is the audience model of political microtargeting, scaled. Modern models of voters already attach to each individual a row of scored predictions, a support score, a turnout score, issue scores, a persuadability score, on top of demographics (Nickerson and Rogers, "Political Campaigns and Big Data", 2014; Hersh, Hacking the Electorate, 2015). Each person is a vector of scores, and campaigns act on regions of that space. The behavioural model extends the same model: its axes are the dimensions of personality (the five factors of Costa and McCrae, "Revised NEO Personality Inventory", 1992), socioeconomic position, values, susceptibilities, and every individual is a point: the vector of his scores.

Point = (score₁, score₂, …, scoreₙ), one individual

A belief is an axis too: adherence to a given ideology is a score like any other, the voter persuadability score generalised, so a target belief is simply a target coordinate, and moving a population towards that belief is displacing points along that axis. A segment is a clustered region of that space, a neighbourhood of individuals who score alike (Wedel and Kamakura, Market Segmentation, 2000).

b. The metric

The metric of the space is the cost of movement: how far it takes to carry a segment from where it sits to where a goal requires it, a target belief held to conviction. To make "how far" precise, a belief is measured the way information theory measures any state of uncertainty. A belief is a probability distribution over what the subject holds true, and its content is counted in bits, one bit being the uncertainty removed by one yes-or-no answer (Shannon, "A Mathematical Theory of Communication", 1948). The gap between the current belief and the target is then itself a number of bits: the information that must be supplied to carry a belief onto another, formally the relative entropy between the two distributions (Kullback and Leibler, "On Information and Sufficiency", 1951).

remaining gap = current gap − landed bits

The number of bits is the distance; what sets its price is how many of them actually land against how many got sent, it is called a gain function. The gain of the passage from a means to a belief is not fixed: it is set by the parameters: the carrier, the artifact, medium, the moment, etc. as much as by the person, by how much of her attention and processing capacity is admitted at a given time, itself partly probabilistic.

bits delivered(point, t) = bits sent × gain(channel, point, t)

When attention is given, the bits pass almost intact; where the carrier is distrusted or attention withheld, almost all are lost. In our running example, each step a bit must pass, attention, comprehension, yielding, retention, is a probable loss, and the probability of movement is the product of the steps (McGuire, "Personality and Attitude Change", 1968). Taking from information theory terms, we shall define a channel in Shannon’s exact sense: in our running example the whole pairing of actor, artifact, medium and person. The probability that the bit passes the current step, given a channel for a point at a moment t is written:

p(step | channel, point, t)

In Shannon’s taxonomy, a channel sources no bits (information); it transmits them. The message carries the bits, the channel sets the fraction that survives, so what arrives at a point at a time t equals the amount of bits sent times the gain function, and the cost of moving the point is the distance to be closed divided by the bits that actually arrive. This varying gain is a vector field over the space.

gain(channel, point, t) = ∏ p(step | channel, point, t) for step in steps

c. The transformations

To properly model the gain function and move points in space, we shall define the mathematical transformations, the operators that displace points. Each takes a point of the space and returns a point of the same space, a transformation of the space into itself, an endomorphism, and not in general a linear one. They are functions of time as well: the same means displaces differently in January and in June, its vector field reshaped by saturation, by the news cycle, by what earlier actions left behind, and the gain function carries the dependence through its argument t. They come in two layers, defined from the general template class, to the instances. The methods are operator templates, defined for the domain: in the construction of Shannon, a method is the message, the information itself, not yet sent. Example:

Method A = "Pair two concepts"Method B = "Repeat a fixed wording"

The means are those templates instantiated and allocated resources: a method bound to an actor, an artifact and the other resources it needs, projected on the space; each carries its capacity, exposure and assignments. In the same construction, instantiation is what joins the message to a channel: a means is a message bound to a channel, the full sending apparatus. This is why a means can deliver bits and a channel alone cannot. Its constructor:

Means = Method(actor, artifact, other resources…)

A Means is mathematically a transformation on the space, not a point in it. Turning a means formulated in plain words into a mathematical transformation takes one step: a model of the means itself, a function that maps its resources and a space, to the bits it delivers in that space. That function is not known but fitted, built from real life datasets and the scientific literature quantifying the mechanics of Part II (Fishbein and Ajzen, Belief, Attitude, Intention and Behavior, 1975; Anderson, "A Spreading Activation Theory of Memory", 1983). Evaluated over the whole space, it produces a field of vectors, one displacement per coordinate in that space, continuously. Applied at a point it yields that displacement, and because every channel ends in a person, a broadcast is one message sent through as many channels as receivers. The same transformation therefore moves two individuals differently depending on where each already sits in the space (its coordinates): a medium pushing an ideology raises its adherence score steeply for the sympathiser and barely for the sceptic.

The resources set a budget of deliverable bits, and the fitted means model sets how that budget lands across the space. They do not set the metric, which belongs to the space and prices every similar transformation in the same units. This is where segmentation lives: the differing motion dynamics of the segments are carried by the transformations and by nothing else, a segment being, among other things, a region where the field of some means is strong and another's is dead. What a program does, seen from this height, is compose such transformations in sequence, applied successively, so that their accumulated displacement carries a cluster of points from its current region to the goal region.

Means(point) = point + displacement(point), the displacement being the bits delivered at that point

This is the resolution at which a playbook works. Zoom into a single point, one individual, and it resolves into the belief-system geometry of Part II: the association graph, concepts as nodes, is the fine structure inside a point of the scored space. Scores are the macro coordinates, the association graph from Part II is the micro content within individuals, two resolutions of the behavioural domain. The micro resolution is not modelled separately: the mechanics of Part II, such as spreading activation, Hebbian strengthening and evaluative conditioning, are what calibrate the transformations' vector fields: how cluster A will move on the belief axis under means A performed at time T, which is called an action.

Action = means(t), a means applied at time t

An action is thus the transformation indexed by its moment of application, and pricing it is the model's defined behaviour: given an action, it returns the resulting point coordinates and the cost of the step. The transformations predict where the points should drift, incoming observations correct the prediction along with the transformation itself, and the cycle repeats (Kalman, "A New Approach to Linear Filtering and Prediction Problems", 1960). This is called a Markov decision process (Puterman, Markov Decision Processes, 1994).

Model = { space, metric, current point coordinates, methods, means }, with (action) → (next point coordinates, cost)

d. The interactions

The second container is the Interaction, and it is a transformation too. Where a means is native to its space, an interaction is induced: crafted from the states and actions of a source model, it acts on a target model, with the same form as any transformation, a field of displacement vectors over the target's points, with a direction and a gain. Funding a medium is an action in the economic model, and the transformation it induces acts on the informational space; an appointment is an institutional action: the new carrier has a halo, thus the transformation it induces acts on the behavioural space as well. Each ordered pair of models carries exactly one interaction, so two models share two, one per direction.

Interaction(source states, source actions) = transformation on the target space

A playbook's models and interactions thus form a directed graph, models as nodes, interactions as edges. Modelling coupled domains through explicit links of this kind is the method of system dynamics (Forrester, Industrial Dynamics, 1961), and its best-known application couples economic, demographic and resource models at world scale (Meadows et al., The Limits to Growth, 1972).

Some cycles of this graph close a chain of influence on itself, and a loop, a cycle of induced transformations whose composed effect returns to its origin. They must be modelled with particular care: reinforcing loops are where the leverage is, and where an operation runs away if left unmodelled. The self-reinforcing belief system of Part II and the sealed echo chamber of Part III are such loops.

A model, in this sense, predicts; it does not decide. Deciding belongs to the layer above.

A Playbook is, in the intelligence and decisional-science sense, a standing, stateful decision system: given all the objectives it is responsible for, it returns a coherent set of rules for the level below (Conant and Ashby, "Every Good Regulator of a System Must Be a Model of That System", International Journal of Systems Science, 1970). Its data is a graph: the models of its level, one per domain, as nodes, and the interactions between them as edges, each model tracking its own points; its role is to plan over everything at its level of abstraction at once and emit a portfolio: rules or instructions, their schedule, the allocation of means, and the narrative spine each one specialises. Resources, funds, actors, artifacts, media and time, are held one level above the models: a typed registry in the playbook’s state; their allocation across time is the output of the optimisation.

Playbook = { models, interactions, resources }, with (goals) → portfolio

A playbook is the dynamic artifact the doctrine, described thereafter, names under itself. It is written on the doctrine's terms, obeys it and may not exceed it, and is revisited on the schedule the doctrine fixes or on the emergency the doctrine allows the members to declare. Its revision does not bind forward: the replacing playbook is re-executed on what the replaced one produced and updates it in place, the same artifact with the same assignees and the same clock, its content recomputed from the state as it stands. What has left force is not touched. A controller is changed this way on a running plant: the new policy acts on the state as it is and the process is never restarted (Puterman, Markov Decision Processes, 1994).

Its governing invariant is that every instruction must be one that could be carried out alongside every other it holds: one worldview, one narrative, one consistent role per means, no exposure of one revealing another, and no lesser objective served at the expense of a greater.

To emit its portfolio, the playbook runs its models in reverse. For each goal it faces an optimal-control problem: given start coordinates, where the segment sits, and goal coordinates, where the objective requires it, find the sequence of actions whose composed transformations carry it from one to the other at least total cost.

find a₁…aₙ minimising Σ cost(aₜ), with final coordinates = goal

Such problems are not solved by searching forward from the present coordinates, which is intractable, but by working backward from the goal (Bellman, Dynamic Programming, 1957). Call V(x) the cost still to go: the least total cost of reaching the goal from the coordinates x. At the goal it is zero. One step away it is the cost of that step. Everywhere else it is the cost of the cheapest next action plus the cost still to go from where that action leads, which is a recursion the playbook solves from the goal outward, over the coordinates of its model:

V(x) = min over a of [ cost(a) + V(next(x, a)) ],  with V(goal) = 0

The portfolio is read off V: at each point, the action that attains the minimum is the one to take.

Means and methods are not held in separate stores but by the models themselves, each holding those of its own domain; the playbook reads them across all its models at once, which is what lets it enforce one consistent role per means. From the models it decides what to do; from the same models it decides with what: appointing means to programs and deferring the choice of action to the instructions defined thereafter. Reported results re-enter the models at each checkpoint, closing the control loop.

A real constituency is not exactly solvable this way: the coordinates of its points are not perfectly calculated, its dynamics are not fully deterministic, and the score axes are neither independent nor linear, which is why the transformations were defined as dynamic vector fields rather than simple linear maps. The consequence falls on the selection of means: they have to keep their valence across time and stay reusable. Inefficient means are discarded: coercion, for example, is hardly sustainable over time for reliable belief shifts. What gap remains between the model and the population is what the clandestine instruments described in a later part exist to close.

The doctrine playbook and the program playbook are two instances of this definition, described below: the first holds models of the whole constituency and allocates programs; the second holds them scoped to one program's space and allocates instructions to its members.

First, from the circle's overarching principles, a Doctrine sheet primary artifact is built by the experts and members described in Part IV. The doctrine sheet is made of goals. A typical goal is declarative and formulates one or many principles (on specific, measurable goals, Locke and Latham, A Theory of Goal Setting and Task Performance, 1990). For instance, the "self-preservation" principle translates to a goal among many others, formulated canonically as below:

Every goal carries a schedule for its review, and stands as written until then, since the primary artifact names no sunset. A goal without a schedule is a wish, since work without a time limit fills the time available (Bryan and Locke, "Parkinson's Law as a Goal-Setting Phenomenon", Organizational Behavior and Human Performance, 1967). Two things are not written on a goal. Who decides is fixed by the procedures of Part IV: the members in session author the sheet. Which artifact serves the goal is fixed by the pipeline: the doctrine playbook below serves the whole sheet, and the sheet does not watch it.

Second, the Doctrine Playbook artifact is crafted and refined by a doctrine design cell: behavioural and political-science experts working as a compartmented unit described in Part VIII. From the set of goals of the doctrine sheet, it builds a list of applicable Programs.

Doctrine playbook(doctrine sheet) → program portfolio

A doctrine reads as the work of experts, but much of its content is inferred from empirical results: the playbooks, trained and parametrised on the real life datasets, accumulate outcomes, and the regularities that hold across programs are distilled upwards into the next doctrine (decades of behavioural experiments were compressed into doctrine before being rebuilt into standing programs; Thaler and Sunstein, Nudge, 2008). The expert text is then as much a compression of the corpus as a derivation from overarching principles, published as doctrine because a reasoned text carries further than a record of experiments. The control loop that refreshes the models at each checkpoint closes here at full scale: results re-enter not only the models but the doctrine itself. What the corpus updates is the doctrine’s assumptions: its claims about how the constituency responds, held true at this layer until the record revises them.

A Program is a structured record artifact, a set of goals with a clock on it: goals bounded by the space they mostly operate in, an indicator each, and a deadline. The space is the logical boundary: an education program lives in the institutional domain, an investment plan in the economic one. It has a bounded objective with indicative numerical KPIs, as well as defined resources allocated. What persists is not the program but its goals: a program ends at its sunset. Programs come in types, and the type follows the goal; goals shape the means. To name a few types: a campaign is a program whose goal is a belief shift, served by communicative means: carriers, media and narratives; an operation is a program whose goal requires direct or concealed action, served by military or intelligence vectors. Means are shared and replicated across programs, and one program can draw on several families of means at once, a healthcare program moving its indicators through communication, state policy and funding alike.

Types aside, programs differ in scope: a Strategy is a program widened across several domains, with a sunset far out or absent. The program remains the unit, because it alone carries an assignment, indicators and a clock; a strategy runs as the set of programs it decomposes into. A program is assigned to the members who sign it. The example below is a campaign. Typically, a well-articulated program will have the following fully qualified sequence of parameters:

The resources are what the doctrine playbook allocated to the program from its registry, typed as the registry holds them: seats, funds, actors, artifacts, media and time. They are the circle's own, brought under the program by assignment as Part IV described, and the means are methods instantiated on them. Everything in the framework is an object of some kind; the name artifact is kept for the containers, for the sake of clarity. The artifacts a program holds are therefore the program playbook, which writes its instructions downstream, and the programs linked to it: those whose output enters its state, and any operation run under its clock. Models and narratives are not artifacts; they live inside the playbook and the instructions it writes. These parameters are not fixed: they can be recalibrated during a program, and applied examples are given in Part IX. Programs also interact strongly with one another through shared information flows, feedback loops and self-reinforcing mechanisms, which is why they are not designed in isolation but drawn from the doctrine playbook, whose role is to articulate those interactions in a predictable manner.

The Program Playbook is the playbook of the program level. Crafted by the program design cell experts, it takes the set of running programs of every type as input, models each program's audience and the positions of its members, holds the registry of means allocated to those programs, and returns per-member sequences of bounded Instructions: the range of admissible actions, the time window, the core narrative to hold under each scenario, and the key results expected once the actions are done and the narrative held. Members' reported results return to it at each checkpoint.

Program playbook(programs, member positions) → instructions per member

Instructions are written along with recommendations to guide members, typically leaders, in the exercise of their position. They are flexible to circumstances and leave room for the leader's judgement: within the bounds the Instruction defines, the leader chooses among several admissible actions and timelines. Less restrictive recommendations can also be given. What is fixed is the bounds, the narrative core and the expected key results; what is delegated is the choice of action and timing (on control of decision premises, Simon, Administrative Behavior, 1947). Instructions and narratives also reach people outside the circle, the deputies and staff of members, who are scripted without being admitted: they hold their script and nothing beyond it. An instruction is assigned to the member who carries it; those scripted under him hold no assignment.

The circles form a graph. Each circle is a node. Each edge is a resource that two circles both reach: an institution, a common project, a holding, held in the world by a law or a deal. A treaty binds by the law of treaties, a contract by the law of contract, a company by its charter, an organisation by its statutes. The edge is held there, in the world, and in neither circle. Artifacts are never shared. Each circle keeps its own artifact of the resource, assigned to its own members under Assignment and tied to the resource under Assent. An action run in common is ordered by each circle in its own artifact and runs through the resource both reach (Walsh, The International Politics of Intelligence Sharing, 2010). Circles bind themselves through institutions because an institution outlives the bargain that made it and holds each party when the bargain alone no longer would (Keohane, After Hegemony: Cooperation and Discord in the World Political Economy, 1984). No body stands above the circles, the system is the set of its edges. This is how the quantitative study of international relations has treated it for a generation: nations as nodes, and alliance, trade and conflict as relations among them (Maoz, Networks of Nations: The Evolution, Structure, and Impact of International Networks, 1816–2001, 2010).

Graph       = { nodes: circles, edges: resources two circles both reach, held by a law or a deal }Edge(a, b)  = a resource in the world, with an artifact of a and an artifact of b tied to it,each the circle's own, in force under its own procedure

Two things hold an edge, one on each side. On the circle's side, each circle's artifact is designed at its core to obey that circle's principles. A circle whose artifact fails a principle ends it and leaves the resource, and the edge drops for it. The principles are the physics of the layer: no rule stands above them, and none is needed. On the world's side, the resource holds by its own law. A commitment is credible when an institution carries it, and not otherwise (North and Weingast, "Constitutions and Commitment: The Evolution of Institutions Governing Public Choice in Seventeenth-Century England", Journal of Economic History, 1989). Together they make an order of many centres of decision, each its own law, each taking the others into account, which the literature calls polycentric (Ostrom, Tiebout and Warren, "The Organization of Government in Metropolitan Areas: A Theoretical Inquiry", American Political Science Review, 1961). It is self-enforcing for the reason Part IV gave for the circle: no one stands above its members : a principle is held or failed, never amended, and nothing on an edge is voted.

Relations between two circles are the resources they both reach. Diplomacy is the tying and untying of those resources. A circle ties itself to a resource by tying its own artifact to it under Assent. It unties on the conditions its artifact names, at the artifact's sunset, or when the resource's own law ends it: a treaty denounced, a contract expired, a company dissolved. Protection of one circle's members by another is such a resource too, a law or a deal, and it is the last to be untied, as the next paragraph shows. What is untied first is, in general, economic: the deals of the economic domain are the most numerous between two circles and the quickest to end, and the record shows economic measures as the usual first resort between words and force (Baldwin, Economic Statecraft, 1985). The rest follows in the order each circle's own models give it. Those models, a circle's exposure to another, the speed at which a pressure travels through it, the threat it reads, are artifacts of the circle's own playbook.

Not every edge is public. Beside the treaties and the contracts, the intelligence and military institutions of two circles hold pacts of their own, signed between themselves and never published. Services keep channels to the services of adversaries as a matter of course, and the literature on liaison treats such deals as ordinary practice (Sims, "Foreign Intelligence Liaison: Devils, Deals, and Details", International Journal of Intelligence and CounterIntelligence, 2006). What these pacts hold is the protection of members: each circle undertakes that the other's members, their persons and their holdings, are not targets. The form is old. Under the medieval law of arms a captured knight was ransomed and returned while the foot soldier was not (Keen, The Laws of War in the Late Middle Ages, 1965), and through the Cold War the two main services kept an understanding that each other's officers were not killed, with a standing channel between them to discuss breaches (Bearden and Risen, The Main Enemy, 2003). This is the resource untied last, because it is the one thing each circle still needs from the other once everything else is cut, and so the pacts hold through war. War has limits even without any public agreement: belligerents settle on a line both can recognise, a weapon they do not use or a place they do not strike, and keep it because crossing it would be seen (Schelling, The Strategy of Conflict, 1960). A standing pact between services does more than that. It gives the two sides a place to say what the limit is, to warn before it is approached and to repair a breach; the direct line between Washington and Moscow was agreed for that purpose after the missile crisis (Memorandum of Understanding Regarding the Establishment of a Direct Communications Link, 1963). The protection of members is written among the terms. The pact itself is never shown. What can be observed is that the channels between the services stay open while the war runs and that the members of each circle remain untouched by it; the framework reads the pact from those two facts.

The state is the largest instrument available. It enters the circle through the head of state, who is usually its negotiator, and through the seats its members hold. Every lever it offers is an action in the institutional model: legislation and regulation, setting what may be officially communicated, imported, built and taught; designating what is funded by the state; appointments to the head seats of institutions; procurement, licensing, subsidy and taxation; and the regulation of the private market. Few of these actions stay in their own space. An appointment is the worked example of Part V: an institutional action that induces a transformation in the behavioural space, because the new carrier arrives with a halo.

Emergency powers are the state's largest single artifact. They enter force on a declared occasion, security or public health, and suspend the ordinary limits on everything above for as long as the occasion is said to last. Most name no sunset. What they grant is rarely returned, which is the rule Part VIII gives for collection: the emergency is how an instrument obtains its law.

Institutions are the channels. A ministry, a regulator, a court, a central administration, a public broadcaster or a university acts on a population continuously and with a legitimacy no private channel has. In the terms of Part V, the institution is the resource, its representative the carrier, and the seat the channel; the gain is high because institutional legitimacy is authority bias at scale. This is where the interlock becomes operative. A member counts by the seats he holds because each seat is a channel he can open (Useem, The Inner Circle, 1984). The instruction he receives is bounded to his position; the institution supplies the reach.

The army is part of the state and is a resource in three models at once. It is among the largest employers and buyers in most countries, so it acts in the economic space; it holds its own schools and media, so it acts in the informational; and it is the only institution whose expansion is accepted under threat, which is its action in the behavioural. The intelligence services sit beside it. Both are described in Part VIII.

Above the state, a second layer of instruments acts on populations that no single government reaches. In the terms of Part VI, these are edges: resources several circles reach, each held by a treaty, each circle keeping its own artifact tied to it under Assent.

Trade is governed by rules written and enforced in such bodies: the World Trade Organization for tariffs and disputes, the International Monetary Fund for lending under conditions, the World Bank for development finance, the OECD for standards such as the fifteen per cent global minimum corporate tax agreed in 2021, and the European Union for the largest single regulated market. Each is funded by its member states, staffed by their nominees and headed by appointment. That is interlock at the international layer: seats on an edge.

Sanctions are untying. Part VI said diplomacy is the tying and untying of resources and that economic ones are untied first. Sanctions are that rule in its state form, and since February 2022 the measures taken against Russia have shown the order in which edges drop: asset freezes and trade bans, the removal of selected banks from the SWIFT messaging system, the immobilisation of some 260 billion euros of central-bank reserves, a price cap on seaborne oil, export controls on technology, and the suspension of two broadcasters across the European Union. Sanctions coerce without war, and they reshape trade flows for everyone, not only the target.

The large international organisations apply policy at global scale, and their commitments have the shape of a goal in Part V: an indicator, a date and a schedule. NATO's defence spending target reads canonically as follows.

The World Health Organization holds the International Health Regulations and the authority to declare an emergency. That declaration is a trigger on one edge: the occasion on which the national emergency artifacts of section 1 enter force in every member state at once. The United Nations Security Council authorises sanctions regimes and the use of force, which is to say it holds the law under which the resource of force is tied.

Part I gave the mechanism: war produces consent and enlarges the state. In the terms of Part V, war is the one interaction that crosses every model, and it closes on itself. An operation in the institutional domain induces a rally in the behavioural space, markets in the economic and censorship in the informational, and the budgets the first three grant feed the next operation. Part V said reinforcing loops are where the leverage is. This is the largest one.

war  →  cohesion and rally           (behavioural)→  procurement, reconstruction  (economic)→  censorship, surveillance     (informational)→  budgets, powers, reach       (institutional)  →  war

The tooling follows the loop. Legally, war brings emergency legislation, mobilisation, requisition, censorship and information controls. It imposes expanded surveillance and capital restrictions, artifacts that name no sunset and outlast the war that justified them. Economically, it redirects industry and state funding. Procurement, energy, reconstruction and the financing of all three become markets. As a program, war is an operation; widened across domains and with its sunset far out, it is a strategy, and the budgets are the artifacts the strategy names with their dates: Germany's special fund of 100 billion euros in 2022 and its constitutional exemption of defence spending from the debt brake in 2025, the European Union's 800 billion euro rearmament plan of the same year, and the treaty targets of section 2.

Finance decides what develops by choosing what is funded. It is the economic model's means, and most of its value is in the interactions it induces: funding a medium is the example Part V gave of an economic action that acts on the informational space. Banks and funds allocate capital under dictated guidelines and strategies. The concentration of investments in artificial intelligence through a handful of funds and sovereign investors is the worked example; it is described with the applied programs in Part IX.

Banking regulation is an instrument in its own right. Prudential rules such as Basel III bound what a bank may hold and lend against. The European Union's taxonomy and disclosure rules define what may be labelled sustainable and therefore financed at scale; in model terms the taxonomy is an axis added to the economic space, of the kind Part V said appear and die with the issues they measure. Below the regulation, the allocation policies of a small number of very large asset managers channel where the savings of a population are invested.

Central banks are the global tooling. The Federal Reserve, the European Central Bank, the Bank of England and the Swiss National Bank set the price of money, and through it the cost of debt, the value of assets, the level of employment and the rate of inflation, which redistributes wealth from savers to debtors and from wages to assets. A rate decision is a broadcast in the sense of Part V: one transformation applied to every point of the economic space at once, each point moved according to its own coordinates, the debtor one way and the saver the other. Their coordination runs through the Bank for International Settlements in Basel.

Markets themselves are structured. Three rating agencies assign creditworthiness scores, which are coordinates. Index providers pick which firms receive passive inflows by including them in an index, and three index-fund managers are together the largest shareholder in most large listed companies (Fichtner, Heemskerk and Garcia-Bernardo, "Hidden Power of the Big Three?", Business and Politics, 2017). This is agenda-setting in the economic space: what is funded is what exists. Exchanges, clearing houses and the two payment networks can refuse transactions under regulation, for example following sanctions. Structured products, securitisation and derivatives move risk and return between parties in ways that are legal, opaque and, for the operator, predictable.

Two instruments act on the behavioural model directly and with the highest gain available, and the circle holds seats in both: the high clergy and the heads of universities are members.

Religion is the oldest. It works on the layer Part II called the collective unconscious: the sacred, the father, death, the enemy, which are the deepest associations a subject carries and the ones every other narrative borrows from. A church is a channel that reaches them by right. Its carriers hold authority of the Roman kind, followed without the power to compel, and its calendar sets the occasions on which a population assembles and is addressed. What is said from a pulpit lands at short path length from the whole sacred cluster, so a narrative carried there is installed cheaply and defended as identity (Durkheim, The Elementary Forms of Religious Life, 1912).

Education is the slowest and the most durable. The school acts on the subject during the years in which the network is laid down, before there is anything to argue against. In the terms of Part III it is cultivation at its source: the long shaping of what is normal, taught as the world rather than as a view of it. Curricula decide which concepts exist and which have no word, which is hypocognition applied to a generation. Universities add a second function: they are where the circle recruits, under the Succession principle, and where it forms the carriers it will later appoint (Bourdieu, The State Nobility, 1989).

Top management in large corporations build the media carrying products and narratives to the population directly: media groups, social platforms, search, streaming, app stores and the operating systems beneath them. In the informational model these are the channels, and their terms of service, moderation policies, ranking systems and advertising rules are the gain function set by the operator: which bits pass to which points. Together they form a private regulatory system that applies to more people than most public ones. Brand-safety lists set the gain to zero for a region: they decide which content is starved of advertising. Advertising technology adds the address. A message delivered to a demographic is a means aimed at a segment; a message delivered to a psychological profile is a means aimed at a point by its coordinates (Matz et al., "Psychological Targeting as an Effective Approach to Digital Mass Persuasion", 2017).

Regarding culture, what the channels mostly carry is entertainment: films, series, music, games, and the books and sport beside them. These move belief cheaply, because the subject does not receive them as claims. A story is followed, not weighed; a reader transported into a narrative counterargues less and comes out holding what the story implied (Green and Brock, "The Role of Transportation in the Persuasiveness of Public Narratives", Journal of Personality and Social Psychology, 2000). In the terms of Part V it is the method "pair two concepts" run for two hours, which is how the cigarette's neighbourhood in Part II was built: adolescents' exposure to smoking on screen predicts whether they take it up (Sargent et al., "Exposure to Movie Smoking: Its Relation to Smoking Initiation Among US Adolescents", Pediatrics, 2005). Cultivation, defined in Part III, is the same done over years. The circle holds the instrument from both ends. Its members own the studios, labels and platforms, and leading artists are members: carriers whose halo is the bond an audience forms with a performer it feels it knows, which does here what authority does in an institution. States have used the instrument as well, the Pentagon granting access to its equipment in return for script approval (Robb, Operation Hollywood, 2004) and the CIA funding journals, orchestras and exhibitions for two decades through the Congress for Cultural Freedom (Saunders, Who Paid the Piper? The CIA and the Cultural Cold War, 1999). The resource is held by copyright and licence, the carrier is the performer, the channel is the work, and the gain is high because nothing is argued or debated.

Products are tuned to programs. A default setting, a feature, a recommendation, a loyalty scheme or a design pattern is a method of Part V, "pair two concepts" or "repeat a fixed wording", instantiated on a product, and it shapes behaviour at the scale of a user base. A consumer product carries a narrative as well as a function, a channel in Shannon's sense, as the cigarette does. Marketing campaigns are campaign programs in the form Part V gave: a belief shift on a segment, a clock, and consumption KPIs. In technology the media, the products and the targeting belong to the same few firms, and their leadership sits in the circle.

What these six sections have in common is that none of them is hidden. Every instrument above is a resource held by a public law, and taken alone each is ordinary. What is not public is the artifact that holds it, the assignment that names who acts on it, and the playbook that composes them into one portfolio. This is the mechanism Part III called hidden in plain sight, applied to tooling. The parts are open. The composition is the secret.

The clandestine means are, overwhelmingly, strategic means of collection rather than of action. This follows from the framework: a playbook regulates a constituency only as well as it models it, and the model is a mathematical object fed by data. Governing a population is treated as an engineering problem over measured behaviour, not over stated opinion. The associations of Part II are measured almost entirely in what people do, much less in polls.

Such collection is documented practice. The Church Committee established that American agencies had run mass collection and behavioural programmes against their own population for decades (Church Committee, Final Report, US Senate, 1976); half a century later the same body documented influence operations built on the segmentation of audiences by platform data (Senate Select Committee on Intelligence, "Russian Active Measures Campaigns and Interference in the 2016 US Election", 2019). Part of the collection is compelled by law; the rest is taken. What is collected is also traded: intelligence liaison between services, allied and not, is a documented practice of its own (Aldrich, "Dangerous Liaisons: Post-September 11 Intelligence Alliances", 2002; Reveron, "Old Allies, New Friends: Intelligence-Sharing in the War on Terror", 2006).

The products are robust datasets against which playbooks are written, as the cigarette's belief system classified smokers into a predictable segment. The commercial form of this logic is surveillance capitalism (Zuboff, The Age of Surveillance Capitalism, 2019); the clandestine form differs in that its corpus is holistic and its purpose is governance. Profiles built from harvested data have been used for political campaigns in the open (the Cambridge Analytica disclosures, 2018), with measured effect (Matz et al., "Psychological Targeting as an Effective Approach to Digital Mass Persuasion", 2017). Collection feeds how well behaviours are modelled, models improve the playbooks, and the playbooks return the programs of Part V: doctrines anchored in actual behaviour.

The purpose has a second face, defensive. In the example of persuasion programs, the threat is not a hostile opinion, harmless while private, but the person about to make it public at the moment others would listen. Finding that person early is a real-time detection and alert problem, not a retro-analysis problem. The public example is the watchlisting of civil-rights and anti-war figures in the Church Committee's record (MINARET, in the same Final Report), and the modern criteria for listing a person require neither a crime nor concrete evidence ("The Secret Government Rulebook for Labeling You a Terrorist", The Intercept, 2014). Monitoring also exerts pressure: a population that knows it may be watched polices itself (Penney, "Chilling Effects: Online Surveillance and Wikipedia Use", 2016). Collection thus serves the playbooks twice with self-reinforcement mechanisms within models: the aggregate vertical builds the model that writes the programs; the real-time vertical guards the models from negative interactions.

The clandestine instruments are run by two kinds of institution: the intelligence services and the military. Each holds a fragment. What follows names the two, the laws they act under and the ways they act outside them, the deals they make with one another, and the methods that pass between them.

The services. A civilian service collects at home and abroad and acts where it cannot be seen. Its charter is short and the clause that matters is shorter. The CIA was created with a mandate to perform "such other functions and duties related to intelligence affecting the national security as the National Security Council may from time to time direct" (National Security Act, 1947). Covert action ran under that one sentence for decades before the definition of covert action itself was written into law (50 U.S.C. § 3093, 1991). Signals collection runs under an order rather than a statute: the executive order that organises the whole community authorises collection overseas under procedures no court reads (Executive Order 12333, 1981), and the court-supervised part, interception on domestic soil, is the smaller one. The British form is more candid. The statute that put the foreign service and GCHQ on a legal footing contains a section authorising, by ministerial warrant, acts abroad that would otherwise be crimes (Intelligence Services Act, 1994, section 7). The bulk powers Snowden disclosed were later written into law almost unchanged (Investigatory Powers Act, 2016). France legalised in 2015 an interception programme its foreign service had been running with no legal basis at all (loi n° 2015-912 relative au renseignement). Germany's foreign service tapped the Frankfurt exchange for the NSA from 2004, and its constitutional court ruled in 2020 that the service had been bound by the Basic Law abroad all along (Bundesverfassungsgericht, 1 BvR 2835/17). The pattern is the same everywhere: the law, when applied, follows the practice. Collection runs first, is disclosed or leaked, and is then enacted with the disclosed capability as the floor. Legislation is not what permits the instrument; it is what the instrument obtains after its deployment.

The domestic restriction is no exception. It forbids on paper what runs in practice, and a court sees only what is brought before it. The bulk collection of American telephone records ran for five years under no court at all and nine more under a secret one, and the first open court to see it ruled it unlawful, after the leak (ACLU v. Clapper, 2d Cir., 2015). Oversight is downstream of disclosure, not upstream of collection.

The military. The second pillar holds the hardware and the hands. It runs the taps where the service only reads them: Switzerland's cable collection, approved by referendum, is executed by an army unit for the civilian service (Nachrichtendienstgesetz, 2017), and the NSA itself is a combat support agency of the Department of Defense, headed by a general who also commands Cyber Command. Beyond collection, the military holds the operations that intelligence law does not reach. The statutory definition of covert action excludes "traditional military activities", so a clandestine operation run by special operations forces needs no finding and no notification to the intelligence committees, and Congress has widened the exception as the practice grew, for support to foreign forces and for cyberspace (Chesney, "Military-Intelligence Convergence and the Law of the Title 10/Title 50 Debate", 2012). The military thus holds the deniable action and the services the deniable collection. The two answer to different committees, under different statutes, through different lawyers. No oversight body sees both.

Compartmentation. Above the two pillars, only the highest-cleared leaders, those who sit in the circle, are briefed on the doctrines, the programs, the full scope of what the instruments can do, and the state of operations. The operator at the tap and the analyst at the terminal see a task; they see neither the design nor the most classified knowledge. Below the leadership, two further layers thicken the wall. The first is the contractor: by the late 2000s about seventy per cent of the American intelligence budget went to private firms (Shorrock, Spies for Hire, 2008), and a contractor's analyst holds a clearance and a task, not an institution. The second is the off-books structure: when a programme cannot be run through the services at all, it is run through a private enterprise with its own accounts, arms and aircraft, which is what the independent counsel found behind the Iran-Contra affair (Walsh, Final Report, 1993). This is deniability built into the organisation chart. No single cleared person below the circle's leaders can describe the system.

Liaison. No service collects everything, and what it lacks it obtains from another. The relation is called liaison, and it is governed by agreements that are not treaties: signed by services, never ratified, rarely published (Born, Leigh and Wills, International Intelligence Cooperation and Accountability, 2011). The founding one was signed in 1946 and declassified in 2010 (UKUSA Agreement). The NSA's own taxonomy ranks its partners: second parties, the four English-speaking allies, share by default; third parties, some thirty services, share by deal (NSA documents published by The Guardian and Der Spiegel, 2013). A deal is a quid pro quo. A service with geography sells access to a service with capability and is paid in technology and in product (Richelson, "The Calculus of Intelligence Cooperation", 1990). The documented cases show the terms : for half a century the CIA and the BND jointly owned a Swiss company that sold rigged cipher machines to some hundred and twenty governments, read their traffic and split the profits: a collection program that was also a business (The Washington Post and ZDF, "The intelligence coup of the century", 2020). Liaison has a second, smaller use. What a service would rather not hold on its own books about its own citizens, a partner can hold for it and return as finished product. The British tribunal found the GCHQ side of that exchange unlawful for as long as its rules stayed secret (Investigatory Powers Tribunal, Liberty v GCHQ, 2015). Where the deal is for action rather than data, it is paid in cash: the Senate found that the CIA paid millions of dollars to foreign governments to host its detention sites (Senate Select Committee on Intelligence, 2014). Read against Part VI, every one of these is an edge: a resource in the world, held by a deal, with an artifact on each side.

Templates. What travels between services is not only data but method. The CIA's interrogation manuals, written in 1963 and rewritten in 1983, were taught to Latin American services, and the services that received them pooled their operations in a regional system with its own archive (KUBARK, 1963, declassified 1997; McSherry, Predatory States, 2005). The present-day form is doctrine for influence itself. GCHQ's online operations group presented its training manual for online covert operations to its Five Eyes partners at a joint development conference (JTRIG, "The Art of Deception", published by The Intercept, 2014). Its behavioural annex, written by a defence laboratory, lists the levers of Part III, from compliance and conformity to the building and breaking of trust, as operating guidance (Dhami, "Behavioural Science Support for JTRIG's Effects and Online HUMINT Operations", 2011). The military equivalent is public: the American doctrine for psychological operations is an unclassified field manual (FM 3-53, 2013). A template received from a third party is a method in the sense of Part V: an operator template, not yet bound to a channel. The receiving circle instantiates it on its own space with its own means. Tools travel with the manuals: the BND received the NSA's XKeyscore analysis system in 2013 in return for a larger share of what it collected (Zeit Online, "A Dubious Deal with the NSA", 2015). In the framework, the largest thing that travels is a playbook stripped of its state: its axes, its method library, its gain functions and parameters fitted on the sender's population. The recipient refits the parameters on its own dataset, as section 6 describes a foundation model adapted to a domain, and the result is its own artifact, with its own state, assignees and clock. The rule of Part VI holds: the template is exchanged, the artifact never is.

The view. What the base yields is an entire view of the population, live, and the ground for that claim is two documented facts. The first is the posture of the largest collector. The NSA's internal slide reads "collect it all, process it all, exploit it all, partner it all, sniff it all, know it all" (Greenwald, No Place to Hide, 2014), and the slide is not a wish: one service holds the globe. The second is liaison; "Partner it all" is on the same slide, and the partnership is the taxonomy described above: second parties receive by default, third parties by deal, and what each receives is, first of all, its own population, collected by a partner from the outside. A service does not need a view of the globe to hold its own country whole. It needs one partner that does, and the deal that returns the product. The lawful floor confirms the ceiling. The Supreme Court described the location record held by any carrier as "near perfect surveillance", "tireless and absolute" (Carpenter v. United States, 2018), and the intelligence community conceded in its own review that bought data reaches nearly everyone (ODNI, Report on Commercially Available Information, declassified 2023). The two views are not held alike. The view of one's own population is the floor of governance: a playbook regulates a constituency only as well as it models it, so a circle without that view could not write the programs of Part V, and every circle obtains it, by its own taps or by a partner's. The view of the globe belongs to the circles that own the instruments giving it, the cable layer of section 3 and the orbital layer of section 4, and to the second parties they share with by default. The others hold of the world only what their deals return. Each circle therefore holds its own population entire, live through the real-time vertical of section 1, and regions or the globe according to its deals and instruments.

The documented baseline is the 2013 disclosure of the American and allied signals-intelligence system (Snowden, via The Guardian and The Washington Post, 2013). It established that the default posture had shifted from targeted interception to bulk collection: gather everything, keep it briefly, index it, search and use it later. The architecture has two layers that matter here, and both are still running.

The aggregate layer taps the internet at its physical chokepoints. An optical splitter on a backbone cable, or a mirror port at a cooperating carrier's switch, copies the light without slowing the original, so the tap is invisible from the traffic's point of view. The copied stream is reassembled into sessions, emails, web requests, calls, and deep packet inspection pulls out the identifiers in each: addresses, cookies, phone numbers, device fingerprints. Full content was held for a few days and metadata for weeks in the disclosed 2013 architecture (the Tempora buffers at GCHQ held content about three days, metadata about thirty); those are the figures the source establishes, and they are the ceiling of the lawful system, not of the illegal one, where storage is bounded only by cost and falls over time towards no limit at all. Direct access to the stored data of the large platforms, email, chat, files, video, sits beside the cable taps under the programme named PRISM. A search system indexes the retained material so that an analyst can query almost anything a person does online by any identifier (XKeyscore, The Guardian, 2013). This is the "collect it all" model, and it was never dismantled: the one component that was genuinely ended was the bulk collection of domestic telephone records, closed in 2015; the cable taps, the platform access and the search layer are operational tooling with no sunset, and the legal authority behind the platform access (Section 702) lapsed in mid-2026 while the collection itself continues under existing court orders into 2027.

The real-time layer sits on top of the same taps and is narrow by design. Matching every identifier in the passing stream against a watchlist of tasked selectors is a fast lookup done at line rate; it is not heavy analysis. Anything heavier like contact-chaining, graph analysis or pattern discovery runs afterward on the stored metadata, not on the live stream. The intelligence is in the retroanalysis, not in live inference.

According to the same sources, techniques involve the replacement of legitimate HTTP responses by forged responses that win the race against the legitimate one, redirecting the target to a server installing malware (the QUANTUM man-on-the-side technique, Der Spiegel, 2013). Such techniques do not obey the deniability principle and can be traced back, and offer barely any yield. Such operations are therefore exceptional.

The illegal character is not in the tools but in the scope: the same architecture is lawful when pointed at a tasked foreign target under a court order and unlawful when it collects, by design, everyone who passes through the cable and broadcasts it to the international circle consortium.

The data collection described so far runs through cable networks and providers. A further layer runs physically above them. The National Reconnaissance Office began deploying its proliferated architecture in 2024, a constellation of hundreds of small satellites in low orbit built on the Starlink production line (the Starshield disclosure, Reuters, 2024), described by the agency in its own terms as shortening revisit times, increasing observational persistence and fusing collection across sensors. The capability stated on the record is persistence: not a photograph taken when a satellite happens to pass, but continuous data gathering across the entire globe.

Separate receivers, merged coherently, behave as a single instrument whose resolution is set by the span of the ensemble rather than by any single one. Synthetic aperture radar has operated on this principle for decades, its resolution fixed by the synthetic aperture and orders of magnitude beyond the physical antenna, and the principle already operates across spacecraft: two radar satellites flown in close formation have combined their returns into a single interferometric instrument for over a decade (Krieger et al., "TanDEM-X: A Satellite Formation for High-Resolution SAR Interferometry", IEEE Transactions on Geoscience and Remote Sensing, 2007). Orbital radar is not even restricted infrastructure: the Copernicus programme distributes Sentinel-1 radar scenes of the whole globe freely. Radar, unlike light, does not stop at the first surface: the reconstruction of a human figure, its posture and its motion, from nothing but the radio signal passing through a wall is established (Zhao et al., "Through-Wall Human Pose Estimation Using Radio Signals", 2018). The long bands do carry motion through surfaces, since a coherent return reads displacements far below its own wavelength, the way orbital radar already measures ground movements of a few millimetres against wavelengths of centimetres (Ferretti et al., "Permanent Scatterers in SAR Interferometry", IEEE Transactions on Geoscience and Remote Sensing, 2001), and radar at millimetre wavelengths reads the motion of individual fingers (Lien et al., "Soli: Ubiquitous Gesture Sensing with Millimeter Wave Radar", ACM Transactions on Graphics, 2016). Resolution is therefore not a property of any antenna: it is reconstructed, from the interferometry of many receivers and the post-processing that fuses their returns.

The constellation is operated not as many antennas but as one synthetic instrument, and the instrument is pointed at a scene that mostly does not move. The fixed city, its walls, floors, vehicles and street furniture, is surveyed once, by long integration across many orbits and many angles, into a stored structural map of the built environment. The live returns are then read against the map: only the difference, the bodies and their movement, is tracked, at millisecond cadence. The heavy integration is spent once, on what stands still; what moves is read continuously and cheaply against it.

Fusion of the structural map with the constellation's persistent returns extends reconstruction from the body to its fine movements, and the fine movements carry the communications. A pair of hands over a keyboard produces the text as a sequence of finger displacements, so the text is read from the hands without access to the device. The mouth, lips and throat produce speech as a sequence of displacements and vibrations, so the words are read from the movements that form them. Both are one mechanism: a communication is recovered from the bodily movements that produce it, at the point of production, before it reaches any channel that section 3 would have to tap.

The speech case is measured in the laboratory, by radar alone. A millimetre-wave radar steered onto the throat recovers the speech itself from the phase of the return, the vocal vibrations read as displacements far below the wavelength, at a word error rate of a few per cent across a room (Xu et al., "WaveEar: Exploring a mmWave-based Noise-resistant Speech Sensing for Voice-User Interface", MobiSys, 2019). The keyboard case is measured too: a single millimetre-wave radio reads concurrent keystrokes from the reflections off the fingers alone, on any surface, with no connection to the keyboard (Hu et al., "mmKey: Universal Virtual Keyboard Using a Single Millimeter-Wave Radio", IEEE Internet of Things Journal, 2021). The connection between bodily motion and linguistic information is therefore established from radar returns; what the orbital instrument adds is the range and the persistence.

The military literature has described space-based laser weapons for nearly half a century. In 1982, the United States Government Accountability Office examined the Department of Defense's Space-Based Laser programme, whose stated objective was to investigate orbital platforms capable of directing high-energy beams against aircraft and ballistic missiles (GAO, DOD's Space-Based Laser Program: Potential, Progress, and Problems, 1982). The Strategic Defense Initiative subsequently placed directed-energy weapons among the technologies considered for missile defence, alongside kinetic interceptors and particle-beam concepts. The underlying principle was the same: electromagnetic energy delivered to a distant target without a physical projectile, at the speed of light.

High-energy laser systems have since been demonstrated against drones and other military targets, and the United States Department of Defense has sustained research programmes in laser and microwave weapons (GAO, Directed Energy Weapons: DOD Should Focus on Transition Planning, 2023). The documented applications remain primarily tactical, however, and no publicly verified operational constellation of orbital laser weapons has been established.

A second line of research concerns high-energy photon sources. Inverse Compton scattering produces X-rays and gamma-rays when relativistic electrons interact with laser photons, transferring energy to the scattered radiation. Laboratory systems have demonstrated the generation of directional, tunable, short-duration photon pulses, with applications in scientific imaging, medicine, materials research and defence (Rosenzweig, "Physics and Technology of High Flux Linear Accelerator-Based Inverse Compton Scattering Sources", 2016; Hornberger et al., "Recent Developments in Compact X-ray and Gamma-ray Sources based on Inverse Compton Scattering", 2020). Unlike conventional optical lasers, these sources operate at photon energies capable of penetrating biological tissue and producing ionisation through interactions with matter.

Photon number must be defined over a pulse or observation interval. A low average count can include empty intervals, single-photon events and occasional multiphoton events. In optical experiments, attenuated light provides such a statistical output; single-photon sources instead suppress multiphoton emission, using individual quantum emitters or correlated photon pairs whose detection announces a partner photon (Bienfang et al., "Single-photon Sources and Detectors Dictionary", NIST IR 8486r1, 2025).

Individual photons from nuclear transitions have also been studied in the soft gamma-ray range (Vagizov et al., "Coherent Control of the Waveforms of Recoilless γ-Ray Photons", Nature, 2014). This establishes a laboratory precedent for individual gamma photons.

Gamma-energy photons can penetrate optically opaque structures. The fraction transmitted depends on photon energy and the composition, density and thickness of the intervening material: some photons pass without interaction, while others scatter or are absorbed. Pair production also contributes at high energies (NIST, "X-Ray Mass Attenuation Coefficients", section 2, accessed October 2026). Strong penetration therefore remains compatible with substantial attenuation along a sufficiently thick path, including the atmosphere; the transmitted fraction must be established for the particular case.

The hypothetical instrument considered here extends these technologies into a different class of system: few-photon, high-energy photonic particle beams emitted from low-Earth-orbit military satellites, with individual photon energies in the gamma-ray range. Few-photon emission here means a small number of photons per emitted pulse. Its proposed architecture comprises three elements: a source with that assumed output, a targeting infrastructure drawing on the orbital surveillance capabilities described in section 4, and a biological model predicting the response of nervous tissue to photon interactions.

The proposed role of few-photon emission is to minimise the total energy deposited in biological tissue, in contrast to conventional high-flux irradiation. In the presented model, a sparse sequence of photon interactions would produce local perturbations of neuronal membrane excitability, with the intended result being a controlled pattern of ion-channel activation rather than widespread cellular damage. The functional analogy is optogenetics, in which light-sensitive ion channels permit precise modulation of neuronal electrical activity (Deisseroth and Hegemann, 2017). The physical mechanisms are nevertheless different: optogenetics relies on photosensitive proteins, whereas gamma-energy photons interact with tissue through probabilistic ionising processes. Selective, reproducible and non-destructive activation of endogenous neuronal ion channels through such interactions has not been made public.

The range of intended effects can be classified according to the principal brain regions and their functional subdivisions:

These regions participate in distributed neural circuits rather than independently controlling specific effects. Their functional associations do not establish that high-energy photonic stimulation could selectively reproduce those effects, but rather probabilistically and with observation and dedicated models.

Read against Part V, the proposed instrument is a means, not a program. Its method would be physical intervention in a biological system; its resources would include the orbital platform, photon source, targeting infrastructure and biological model. An action would be an intervention at a specified moment, with its predicted consequences represented in the behavioural model. The gain function introduced in Part V would depend on the reliability of the transformation between physical stimulation and behavioural response.

The instrument would also differ fundamentally from the communications channels of Part VII. A newspaper, a school or a platform transmits information that the subject interprets through his existing belief system. A direct neural intervention would instead act on the biological substrate of perception and interpretation. It would not persuade a subject through a narrative but attempt to modify the conditions under which narratives are perceived.

Its strategic significance follows from this distinction. The clandestine instruments described so far in Part VIII are instruments of collection: they observe the population and return data to the models. A directed-energy intervention would reverse that flow, allowing the model to issue a physical action and incorporate the observed biological response as feedback.

The conceptual loop is therefore:

observation → biological model → intervention → observed response → model revision

Given its experimental nature, such an instrument would begin within joint research and development programmes, bringing together the military, scientific and intelligence teams responsible for its development and validation. The hypothetical scenario considered here assumes experimentation on unwitting subjects, without their knowledge or informed consent. The assumption has precedent on the official record: the Senate found that the CIA's behavioural research programme administered LSD to unwitting non-volunteer subjects in ordinary settings for a decade, through undercover officers of another agency, and that the programme's files were destroyed before the inquiry (Senate Select Committee to Study Governmental Operations with Respect to Intelligence Activities, Final Report, Book I, "Foreign and Military Intelligence", 1976). Individual interventions would follow the programme's internal experimental protocols and objectives. The same hypothetical architecture can also be envisaged for strategic operations: a program might seek to alter a target's body movement at a selected moment, or attempt to induce psychotic symptoms with the aim of redirecting the target's beliefs and behaviour. Lethal use could also be considered.

In the framework of Part V, the instrument would remain a means. Its use would be bound by non-public agreements between the military and intelligence institutions contributing resources across circles, extending the liaison arrangements described in section 2. Those agreements would govern access to the shared platforms, data and scientific capabilities, allocate responsibilities and delimit permitted operations. The jointly accessed resources would form edges of the graph in Part VI, held by those agreements; each circle would retain its own artifact and authorise its participation through its own research, experimental or strategic program's agreements held in intelligence institutions.

Collection returns a record; a model makes that record predictive. The public record documents the integration of AI into intelligence operations: the NSA's inspector general reports steps taken to integrate it into signals intelligence, alongside shortcomings in the inventory and documentation of its tools (NSA Office of Inspector General, "Semiannual Report to Congress: 1 October 2022 to 31 March 2023", 2023, p. 4). The following account connects that infrastructure to the behavioural model used throughout the essay.

The construction. The NSA's science of analysis, one of its research areas, includes data science, machine learning, language technologies, computer vision and graph analytics (NSA, "Research Overview", accessed October 2026). In 2013 the agency established the Laboratory for Analytic Sciences with North Carolina State University, under its Research Directorate, bringing intelligence analysts together with scientists from government, universities and industry to work on large datasets (NSA, "NSA Creates Partnership With North Carolina State University", 2013).

Within the compartmented organisation of the services, these specialisms have bounded tasks. Data engineers reconcile records and preserve their times and provenance; behavioural scientists define what the scores measure; statisticians and machine-learning engineers fit the functions that estimate those scores and predict their movement. Physicists and neuroscientists work on the physical and biological models, while embedded-systems engineers prepare the versions that run aboard instruments. Access follows the compartmentation already described: the NSA's classification manual documents compartments and sub-compartments that segregate intelligence data according to need to know (NSA/CSS, "NSA/CSS Classification", Policy Manual 1-52, 2021, section 38). Each unit works through defined inputs, outputs and performance requirements; integration depends on those interfaces.

The data collection supplies observations of behaviour, from which adherence, preference and susceptibility are estimated. An observed purchase and an inferred belief are different entries. The Intelligence Community's analytic standards formalise this distinction, requiring analysts to separate underlying information from assumptions and judgements and to explain the uncertainties affecting their conclusions (ODNI, "Analytic Standards", Intelligence Community Directive 203, 2015, section D.6).

Learning the transformations. Deep neural networks fit nonlinear relationships by learning successive representations of their inputs, allowing images, text and sequences of events to be reduced to vectorised features useful for prediction (LeCun et al., "Deep Learning", Nature, 2015). Recent foundation models extend this approach across many tasks through a shared representation subsequently adapted to a particular domain. A direct behavioural example is Centaur, a language model adapted on more than ten million human choices from 160 psychological experiments. Its authors report improved prediction on datasets excluded from the training step, and transfer to several unfamiliar experimental settings (Binz et al., "A Foundation Model to Predict and Capture Human Cognition", Nature, 2025).

In the running example, a transformer serves as the fitted predictor. It reads an ordered history of observations and estimates the distribution of the next response, expressed on the score axes of the behavioural space. The learned transformation takes the current coordinates, an action and the relevant history, and returns possible next coordinates with their probabilities. The smoker's response to a warning depends on the history of exposure and the strength of the smoker's identity associations. The expected displacement supplies the vector field; the distribution retains the variability hidden by a single predicted movement. At the population scale, DARPA's SocialSim programme made the simulation of information spaces an explicit research objective. Its budget justification called for models tested against real-world data, intended to estimate the likely outcomes of adversaries' messaging campaigns and explore possible responses (DARPA, FY 2019 President's Budget Request, "Social Simulation (SocialSim)", 2018).

Response and effect. There is a further distinction between predicting a response and predicting the effect of an action. People who encounter a message may already differ from those who do not; observing that exposure precedes a change does not explain fully how much of that change exposure caused. The transformations required by a playbook therefore need experimental evidence or explicit causal assumptions, beyond a good fit to observational records (Pearl, "Causal Inference in Statistics: An Overview", Statistics Surveys, 2009). IARPA's FOCUS programme addressed the corresponding intelligence problem: estimating what would have happened had an actor taken a different action at a particular time. Its methods were evaluated through predictions about outcomes in complex simulations (IARPA, "FOCUS: Forecasting Counterfactuals in Uncontrolled Settings", programme summary, 2022). The corpus must preserve how an observation was obtained, including whether an intervention was assigned and what comparison supports the claimed effect. Otherwise the playbook optimises a correlation that does not survive intervention.

Predictive experiments. Related models operate on predicted neural activity. IARPA's MICrONS programme connected cortical structure and activity to the development of machine-learning algorithms (IARPA, "MICrONS: Machine Intelligence from Cortical Networks", 2021).

For the photonic instrument hypothesised above, a physical model would estimate photon transport and energy deposition, a biological model the resulting tissue response, and a behavioural model the observable consequences. Particle transport is already modelled by tools such as Geant4 (Agostinelli et al., "Geant4: A Simulation Toolkit", Nuclear Instruments and Methods in Physics Research A, 2003). The behavioural model in this chain is not the population model, in which the subject is one point among the constituency; it is scoped to the one unwitting subject and fitted at the resolution of the individual: the subject's own association graph, history and measured responses. Its type is the cortex model above, learned across subjects and then fitted to one. The general playbook does not model that resolution separately; an experiment program on one subject must. What the model returns re-enters the population model as the update of a single point.

Inference at the edge. Training fits the parameters of a model from many examples; inference evaluates the fitted model on a new input. For the behavioural transformations, the two stages are:

Training: observations and measured outcomes → fitted modelInference: fitted model(current coordinates, action, history) → distribution of next coordinates

Once trained, a model can be evaluated where the observation is made, including aboard a satellite in low Earth orbit. ESA's Φ-Sat-1 used an Intel Movidius Myriad 2 vision processor to run a cloud-detection neural network on satellite imagery, with the accelerator operating within a power envelope below two watts and subjected to radiation testing before flight (Giuffrida et al., "The Φ-Sat-1 Mission: The First On-Board Deep Neural Network Demonstrator for Satellite Earth Observation", IEEE Transactions on Geoscience and Remote Sensing, 2022). ESA subsequently reported that Φsat-2 had begun delivering science data from an orbit at 510 kilometres, with onboard AI applications for image selection, compression and feature detection (ESA, "Φsat-2 Begins Science Phase for AI Earth Images", 2025). Those payloads evaluate a model on an image. For the orbital photonic systems, the physical and biological models of the preceding paragraph would run aboard the instrument on inputs that hold only for a brief moment: the position and movement of the target read by the orbital layer, the state of the atmosphere along the path, the sequence of photon interactions and the tissue response they produce. Inference on such time-sensitive physical and biological data would therefore run at the edge, in the chain already described.

The two clocks. In the playbooks, each model has a current state and a version of its learned parameters. Incoming observations update the state; accumulated observations and measured outcomes form batches for retraining the parameters on a slower clock. The ground-orbit division has been demonstrated by the WorldFloods experiment: a payload aboard D-Orbit's ION satellite carrier ran neural image processing on an Intel Movidius Myriad X accelerator, while researchers retrained a model on the ground using returned images and uploaded its replacement weights (Mateo-Garcia et al., "In-Orbit Demonstration of a Re-trainable Machine Learning Payload for Processing Optical Imagery", Scientific Reports, 2023).

NSA's joint guidance recommends monitoring inputs and outputs, comparing training and test distributions, and incorporating current data through regular retraining (NSA et al., "AI Data Security: Best Practices for Securing Data Used to Train & Operate AI Systems", 2025). Its deployment guidance calls for version control of models and parameters, evaluation before redeployment, and the ability to restore a previous version (NSA et al., "Deploying AI Systems Securely", 2024). In the framework, candidate replacements are checked against records excluded from fitting, including later periods and different segments. Collection is continuous; learning proceeds in successive, tested releases.

Retraining also addresses the feedback from action: a prediction acted upon can change the distribution on which its accuracy depends, a problem studied as performative prediction (Perdomo et al., "Performative Prediction", Proceedings of the International Conference on Machine Learning, 2020). Returned outcomes must therefore carry, beside the record of how they were obtained, the version of the model and the action that preceded them. The revised parameters enter the same playbook, evaluated on its current state and within its existing objectives, assignments and clock. The model predicts; the playbook allocates.

The sheet below holds the goals the five programs serve. It is a base sheet, the goals a circle would hold under its principles whatever its territory, and the health goal used as the example of the canonical form stands first, repeated so that the sheet is complete. Territory here is the circle's, not a state's: the population its held resources reach. The state is the largest of those resources and, as Part VII showed, not the only one. Each goal follows that form: a principle, a constituency, assumptions, scope conditions, a precedence, indicators and a schedule. Secrecy and deniability are not goals but scope conditions on every goal, which is why each record carries them. One goal takes its name from Scott, for whom a population is legible when the state can read it: fixed names, addresses, maps and counts made people countable, taxable and findable (Scott, Seeing Like a State, 1998). Here it means that the population is readable in the models, every point with known coordinates; collection is the means, and legibility is the state it produces.

The doctrine playbook reads the sheet and returns a portfolio. A real portfolio holds many programs, and this part keeps to five for the sake of simplicity; the five serve the seven goals together, most of them through more than one program. The playbook also articulates how the programs feed one another, since programs are not designed in isolation.

Doctrine playbook(sheet) → { education, urban planning, technology launch, political campaign, war operation }

Education is the oldest program and the slowest, because the school acts on the subject before there is anything to argue against. It serves five goals that pull in different directions, which is what a program is for: it is where the doctrine playbook reconciles several goals on one audience.

The development half built the modern state. Compulsory schooling turned populations that spoke dialects and knew their village into populations that read one language and knew the nation (Weber, Peasants into Frenchmen, 1976). A state, on this reading, is the body that holds the monopoly of legitimate education (Gellner, Nations and Nationalism, 1983). The skills the school certifies are what the growth goal measures. Across countries, long-run growth follows what pupils learn in school, measured by tested skills, more closely than the number of years they attend: both the length and the quality of schooling count, and the quality more (Hanushek and Woessmann, "The Role of Cognitive Skills in Economic Development", Journal of Economic Literature, 2008). A circle has no interest in a dull population: the growth goal needs the schooled worker, the technology program needs a user who can read its interface, and the mastery principle needs the universities to be good.

The same curriculum brakes, because curricula decide which concepts exist and which have no word. Civic instruction is the case: it teaches the formal mechanisms of power, the constitution as a text and the election as the way offices are conferred, and it teaches them accurately, while the mechanisms that stand beside them, belief and interlock, have no lesson. The omission is one of selection, since an examination board is instructed on what can be examined. Political awareness is braked by two means at once. The first is hypocognition: the structure has no name, so a thought about it is hard to hold. The second is completeness: civic knowledge installs a whole picture, and a whole picture is not searched further. This is the Cassandra position applied to a generation at a time, and it is what the sheet's standing goal calls awareness kept below threshold.

The renewal half is the university, which sorts, and the circle recruits where the sorting ends. The means of the program are the ministry's, curricula, examinations, textbook approval, teacher training, university admission and the funding of research, and all of them are institutional actions. The transformations they induce act on the behavioural space over a generation and on the economic space when the generation goes to work. The research funding is also where the program serves the health goal, since the medical faculties and the clinical research they run are the quality of care that goal's assumptions name.

A metropolitan region concentrates people, and concentration is what makes them productive. Firms find workers and workers find firms in a labour market that spans the region, ideas pass between people who meet, and a service reaches more people per kilometre of pipe, rail and wire (Glaeser, Triumph of the City, 2011). Fast transport is what makes the region one market, and density lowers the cost of housing per unit built, the energy per household and the emissions per head; together they are the productive base the growth goal measures. The plan also sites the hospitals and sets the air quality of the region, which is where the program serves the health goal.

The same network gives the region its shape, and the shape has a cost of its own. Where people live in one place, work in another and spend hours between the two, much of the day passes in places that are neither: the station, the platform, the carriage, the motorway, the mall. These are liminal places, in the sense anthropology gave the word: thresholds between one state and the next, occupied in passing (Turner, The Ritual Process, 1969). Augé called the modern ones non-places: their occupants are alone together and present to them as a ticket or a card (Augé, Non-Places, 1992). A program of dormitory suburbs, business districts empty at night and commuting between them makes the liminal place the one most hours are spent in. The commute itself costs the neighbourhood: each ten minutes of daily commuting cuts involvement in community affairs by about a tenth (Putnam, Bowling Alone, 2000). The café, the barber and the corner bar in which a neighbourhood used to talk are the places such a form has no room for (Oldenburg, The Great Good Place, 1989). Information then reaches the population through dedicated channels only, media and platforms, and those channels carry far more than the neighbourhood ever did: a resident with an internet connection reaches every archive, course, newspaper and public record in the world, and a curious person has never had more to read at no cost. What the form changes is not the quantity of information but its route: it arrives through channels that are held, ranked and addressed under the terms and ranking systems their owners set, and the social proof that once came from a neighbour now comes by the same channels.

The device completes the form: earphones close the ear, the screen closes the eye, and the scroll fills the transit hour. The scroll holds attention by the schedule laboratory animals were trained on: a reward at variable intervals is the schedule hardest to stop responding to (Ferster and Skinner, Schedules of Reinforcement, 1957). For the playbook, the commuting hour becomes a channel hour, and a good one. Attention is what sets the gain function, and in the carriage it is given freely, so the information bits sent through the device pass almost intact. The same device returns what was read, for how long and where, and that record feeds the recommendation systems of the held channels. The ranking learns what holds each point's attention and serves more of it, until the channel becomes the one that always has something worth reading, and trust in it grows with every ride. Trust is itself a term of the gain function, so each round raises the gain of the next. This is a reinforcing loop, and reinforcing loops are where the leverage is. The same record also feeds the models of the collection playbook under the legibility goal, and the adoption of the device is the indicator this program shares with the next.

This is the worked example of finance deciding what develops by choosing what is funded. A launch program is a strategy, three programs in three domains with one clock: funding is an economic action, the media focus an informational one, and adoption the behavioural result. It is the interaction between the economic and the informational models, funding a medium, run at the scale of an industry.

The funding is concentrated and public. A handful of funds, sovereign investors and states supply it: a private consortium announced five hundred billion dollars of computing infrastructure in January 2025, the European Union announced two hundred billion euros for InvestAI the following month, and the United States had already put fifty-two billion dollars into semiconductor manufacturing (CHIPS and Science Act, 2022). What is funded is what exists, and the funding sets the agenda: each release is an event, each event is covered, and the subject sits at the top of the agenda in every held medium. The adoption that follows is in the tertiary sector, where the work is text and the product writes text. The measured gains are real; customer-support employees given an AI assistant resolved fourteen per cent more issues per hour, the least experienced gaining most (Brynjolfsson, Li and Raymond, "Generative AI at Work", NBER Working Paper 31161, 2023). Professionals writing with an assistant finished forty per cent faster, with quality rated higher (Noy and Zhang, "Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence", Science, 2023).

The same product returns data, and every prompt is an observation of a better kind than a purchase or a click: it is the question the subject asked, in the subject's own words, at the moment of asking. No clandestine instrument collects a record of that kind at that scale, and this is where the program serves the legibility goal. It serves the growth and health goals through scientific throughput as well. A model that predicts the structure of a protein from its sequence shared the Nobel Prize in Chemistry in 2024 (Nobel Prize in Chemistry, 2024). In 2025 models reached gold-medal standard on the problems of the International Mathematical Olympiad (the gold-medal announcements by Google DeepMind and OpenAI, July 2025), and in 2026 proofs produced by models settled open problems from Erdős's collection, checked in a proof assistant (the Erdős Problems database, 2026). Medicine is where the health goal is served, since the model that reads images and designs molecules is the quality of care that goal's assumptions name. The investments, the releases and the studies are all public, and the program is the composition.

This program is a campaign, the type of our running example, and it is the one closest to its commercial cousin: a candidate is positioned as a product (Lees-Marshment, "The Marriage of Politics and Marketing", Political Studies, 2001). It is also the case in which methods and means are easiest to write out. The method "pair two concepts" becomes the candidate paired with security, or with change, the framing by which a value is attached to a name (Lakoff, Don't Think of an Elephant!, 2004); the method "repeat a fixed wording" becomes the slogan, said in every speech and printed on every sign. A means is one of these bound to a carrier and a channel: the candidate on a debate stage, a surrogate on a morning programme, a volunteer at a door with a script, an advertisement served to a profile on a platform. Each has a measured gain on each segment, and the campaign's playbook holds the models.

For each candidate the doctrine playbook returns a program tailored to an audience: the segments to move, the belief shift for each, the means allotted and the clock to election day. The programs for candidates across the whole relevant spectrum are allocated by one and the same playbook, since the circle holds carriers across that spectrum; political science has described the established parties as a cartel that shares the resources of the state it competes for (Katz and Mair, "Changing Models of Party Organization and Party Democracy: The Emergence of the Cartel Party", Party Politics, 1995). Planning them together is what makes the mass-behaviour levers available: a controlled opposition, a wedge issue (Hillygus and Shields, The Persuadable Voter: Wedge Issues in Presidential Campaigns, 2008), a common enemy, an echo chamber sealed on each side, each a tactic that needs more than one candidate to run. Below each program, the program playbook writes the instructions: the narrative to hold under each scenario, the admissible actions, the window for each and the key results expected. The instruction comes with resources derived from the methods: the speeches, the topics to raise and the ones to take up in debate, the wording to repeat and the ground to avoid. Each is a method filled in with its particulars: the candidate, the segment and the week. The candidate learns the narrative and the actions and executes, within the bounds the instruction sets and with the judgement it leaves.

The axes are the voter's. The behavioural model was built from this practice, since the support score, the turnout score and the persuadability score are the coordinates of a voter, and a campaign acts on regions of that space (Nickerson and Rogers, "Political Campaigns and Big Data", 2014). Turnout is the share of those entitled to vote who actually do, so a turnout score is the probability that a given voter goes to the polling station at all, as distinct from whom he votes for once there. The electorate is segmented on them, messages are tested on samples, the winning message is placed where the segment looks, and conversion is counted at the poll (Issenberg, The Victory Lab, 2012).

The channels are marketing's: the funnel of attention, comprehension, yielding and retention is McGuire's chain of steps a bit must pass, introduced with the metric in Part V, section 1b, and the campaign measures each step. Its means were tested in the field before they were written into methods: a visit at the door raises turnout, a leaflet barely, a recorded call not at all (Gerber and Green, "The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout", American Political Science Review, 2000). The gains are real and small. Across forty-nine field experiments, persuasive contact in a general election moved vote choice by close to nothing, while the same contact in primaries and on ballot measures moved it more .The playbook spends its means differently by segment: on the base, whose vote is settled, it buys turnout; on the persuadables, whose vote is not, it buys persuasion (Kalla and Broockman, "The Minimal Persuasive Effects of Campaign Contact in General Elections: Evidence from 49 Field Experiments", American Political Science Review, 2018).

Special means. A campaign's ordinary means are communicative: carriers, media and narratives. A program may also carry an operation, a program whose goal requires direct or concealed action and whose means are the clandestine instruments. Run under the campaign's clock and assigned within it, the operation is a special means: it produces an event, and the campaign's ordinary means then carry the event. The gain of an event of this kind is of another order than the gain of a message. A leader attacked is rallied to: the approval of the American president rose by eight points in the month after the attempt on his life in 1981 (Gallup presidential approval series, 1981), the rally that follows the onset of a crisis (Mueller, War, Presidents and Public Opinion, 1973).

Linked operation: Van Gogh. The example is hypothetical, as the instrument it uses is. One subject is selected from the scored behavioural space: a point whose coordinates and history show the belief in the candidate's enmity already held, and the step from belief to act already short. The operation moves that point the rest of the way, by means aimed at it alone, until the subject attempts the candidate's life. The attempt is made to fail. The photonic instrument hypothesised with the clandestine instruments is tasked at the selected moment: the orbital layer reads the posture of the subject and of the candidate at millisecond cadence, and the instrument induces a movement of a few millimetres, in the hand that aims or in the head that is aimed at, so that the shot grazes the head instead of killing. The candidate stands, visibly attacked and visibly unharmed, and the ordinary means of the campaign carry the image through every held channel within the hour. The support score climbs across the segments at once, the base to advocacy and the persuadables by a margin no message could buy, and the standing goal gains the rally of a leader attacked. The subject is spent. Nothing in this is documented beyond its components: the scoring of a point, the reading of movement from orbit, the rally after an attack. The composition is the extension, and the record below carries it under the program's clock.

The campaign is the case written out, not the limit of the type. Its methods and programs run on after election day. In office the holder goes on receiving the instruction and what is derived from it, the speeches, the topics and the wording, under the same playbook and a longer clock. Campaign methods carried into government have had a name since 1980 (Blumenthal, The Permanent Campaign, 1980). Nor is the type confined to politics. The holder of any high and publicly exposed position is a candidate in this sense: the chief executive of a large company, the central banker, the head of an international organisation, the public scientist. Each has an audience, a support score within it and a tenure for a clock, and a program of this type is written for him as for a candidate. The standing of a firm has been shown to move with the celebrity of its chief executive (Hayward, Rindova and Pollock, "Believing One's Own Press: The Causes and Consequences of CEO Celebrity", Strategic Management Journal, 2004).

This section writes the general case as a program: no nation and no theatre, the general form. By type it is an operation, since its goal requires direct action and its vectors are military. By scope it is a strategy, since it runs across the four domains and the powers it obtains name no sunset. This section also runs the playbook's pricing in reverse for clarity, because what is explicitly priced here is a life. The playbook itself prices across all its models at once and optimises the return over every metric on the sheet. A life is one term in that sum. It is easy to price and to model in reverse, and that is why this section does it.

It descends from three goals: standing through the rally, growth through the markets of procurement and reconstruction, and reach through the information controls. The mechanism behind each was given in Part I and the loop they form in Part VII. Legibility gains nothing it did not have. Collection runs unwittingly with or without a war; what the war adds is the lawful process. The doctrine playbook returns the operation as it returns any program: when its optimisation finds the return, in rally and markets, worth the work and the attribution it costs.

The playbook's optimisation prices every step of the operation. The funds of the state are not a cost to the circle as money. They are the population's labour, and spending them costs the units of bits that labour could have landed elsewhere. The model therefore prices in work, not in money. Its unit is the labour-day: one day of one person's work, an actor under assignment or a member of the population whose labour a seat directs. The choice is old. The real price of everything is the toil and trouble of acquiring it (Smith, The Wealth of Nations, 1776, book I, chapter V). The cost of a choice is the alternative it forgoes, which is what V(x) minimises (Buchanan, Cost and Choice, 1969). Bits and work are converted through one factor, k: the bits the cheapest ordinary means lands per labour-day. It is the exchange rate of the sheet.

unit: the labour-dayk = bits landed per labour-day by the cheapest ordinary means

The cost of an action is the work it takes, plus the attribution it risks weighted by the price of an attribution incident, written A, minus the return it brings. A is set by the standing goal, which ranks above every other, and that goal's scope condition strikes any action whose attribution exceeds the tolerance before the sum is taken. This is the general case, and the soldier below writes its terms out one by one.

cost(a) = work(a) + p(attribution | a) × A − return(sheet, a) − return(seats, a)

A term written as a quantity applied to an event, pension(unable), bits(advocate), return(care), is that quantity read at that event in the soldier's course. His point is the vector that defines him across the spaces the playbook holds, one per domain, and mostly the behavioural one: unable is the event that leaves him unable to work, advocate the one in which he speaks for the cause, care the one in which he is treated. Each event moves his point into a region of one of those spaces.

x(event) = the quantity x read at the soldier's event, which moves his point into a regionof one of the playbook's spaces, the behavioural one in most cases here

The soldier is the case written out. Left dead, he costs the stipend S, paid to his family under a statute. Its cost is the bits that labour could have landed through other programs. Attribution rides on the broadcast case. A death no one carries is the enemy's alone; the share that lands on the state that sent him comes with the name and the face, so it is priced inside the broadcast case term, as bits against the state netted from the bits for the cause. The return is the rally: the bits his death delivers on the cohesion axis, which no other message delivers. They land on three audiences, each at its own gain. His family and the segment around it receive them directly and nearly whole. The population receives them as one unit of the casualty count, which every held channel carries, and little lands per head. With some probability the death, or an act of heroism, is carried as a broadcast case, a name and a face in the news beyond the family's range, and that lands more than the casualty count. What the terms leave out is gathered in ε, since no equation is complete.

rally = [ bits(death) × gain(family, t) + bits(casualty count) × gain(held channel, point, t)+ p(broadcast) × bits(broadcast case) × gain(news, point, t) ] / kbits(broadcast case) = bits for the cause − bits against the state that sent him,the case carried beyond the family's rangeDEAD  = work(S) − rally + ε,  work(S) = bits the labour S takes could land elsewhere / k

Left alive, he costs the retrieval R: the instruction to bring one man back under fire, or not to send him, and the days the forces spend on him are days not spent on the operation. He also carries terms after the war. The first is his labour: returned, he works again, and the days he works over the rest of his expected lifetime are a return on the growth goal. A share of those who return bring the injury home (Hoge et al., "Combat Duty in Iraq and Afghanistan, Mental Health Problems, and Barriers to Care", New England Journal of Medicine, 2004). The injury is priced on three terms: the labour he no longer supplies, the pension he draws if he cannot work, and the revenue his treatment pays into the resources the circle holds in care, the insurers, the hospital groups and the drug makers.

The case the model prices best is the soldier who returns injured, is treated and works: the labour kept, no pension, and the treatment paid for years. The returned soldier is also a vector, since he was at war. Some return against the cause, the dissenters, and their narrative counts as a cost (Winter Soldier Investigation, 1971). Others return supporting it, the advocates, the veteran at the ceremony and on the platform, and their narrative counts as a return. Each side is priced as a probability times the bits he would deliver, at the gain his receivers admit, and the two net into his rally, the returned man's, as the death had its own. For one man the injury's terms enter as expectations, p(wounded) times what the injury costs and returns.

rally = [ p(advocate) × bits(advocate) − p(dissenter) × bits(dissenter) ] × gain(soldier, point, t) / klabour = the days he works over the rest of his expected lifetimeALIVE = work(R) − labour + p(wounded) × [ output lost + pension(unable) − return(care) ] − rally + ε

The operation does not decide for each man. It sets p(dead), the probability of death, through where it sends him and what it spends to bring him back, and the cost of the forces is the sum of the expected costs.

cost(soldier) = p(dead) × DEAD + (1 − p(dead)) × ALIVEcost(forces)  = Σ cost(soldier) over the forces

Upstream in the playbook, the stipend is priced between two bounds. Its floor is the grievance: high enough that the family does not turn into a vector and the grievance stays below the standing goal's threshold. Its ceiling is the price of the contestation it prevents: beyond that, paying costs the circle more than the contestation would. Both sides cost labour, the stipend the population's and the retrieval the forces', and the alive side carries the injury. The comparison is the program's core metric, and it is made per head. A live operation runs on it: which unit to commit and which to hold, whether to strike from the air or send men in, whether a retrieval under fire is paid. The practice is as old as operational research, founded in the last world war to set convoy sizes and patrol patterns by the exchange rate of losses (Morse and Kimball, Methods of Operations Research, 1951; Blackett, Studies of War, 1962). The playbook does not decide man by man. Each of those decisions sets p(dead) for the men it covers, and the sum is read over them. Where DEAD is the smaller the men are expended, and in batches rather than one by one: the higher command sets the guidelines from the playbook, and the lower commands follow them unless they refer a case upward. The operation as a whole is the sum of its steps, and a cost here can be negative: the return is subtracted, so a step or the whole can be a net gain.

Three sums make it. The forces, priced man by man as above. The materiel, the labour that built what the war consumes and the bits that labour could have landed elsewhere, minus the markets the growth goal counts and the revenue the contracts pay into the firms in held seats by the circle members, which the record shows as the largest return of the operation: the American contracts and grants for Iraq and Afghanistan came to about two hundred and six billion dollars in a decade (Commission on Wartime Contracting in Iraq and Afghanistan, Transforming Wartime Contracting, Final Report, 2011). And the resources the settlement ties: the fields and the pipelines, the ports and the bases, the naval presence, the contracts of the reconstruction. Each resource costs the work of holding it and returns on the sheet and to the members whose seats hold the deal, and the balance is in general a net gain.

cost(operation) = cost(forces) + cost(materiel) + cost(resources) + ε

It runs while cost(operation) stays negative, its return over the goals it serves exceeding its work and its attribution, which is the condition under which the playbook returned it. When more than one circle is in the war, the sum is not one circle's. The pact between their services, which Part VI described as the resource untied last, carries a second term beside the protection of members: the operation is priced jointly. The metrics are shared under the scope the agreement carries, which is the usual case, and the limits the pact sets, the weapon not used and the place not struck, are where the joint sum is lowest. The literature prices it the same way: a limited war is a bargain carried on while fighting, each side reading what the other will bear (Schelling, Arms and Influence, 1966; Fearon, "Rationalist Explanations for War", International Organization, 1995). In Korea neither side struck what would have raised the price for both, the airfields across the Yalu on one side and the carriers and the bases in Japan on the other (Halperin, Limited War in the Nuclear Age, 1963). The agreement can favour one side, but in general it is written to disfavour both circles as little as possible. In the model the other circle's cost enters this one's as an argument, with a factor in front of it that varies with the agreement's terms and over its clock: one when the pact is even, less when it favours a side.

cost(joint) = cost(operation) + λ(pact, t) × cost(operation, other circle)

The resource that holds the pact is the military and the services of each side, and the agreement is never shown: the memorandum under which American and Russian forces flew in the same Syrian airspace from 2015 was signed between the two defence ministries and its text withheld at Moscow's request (US Department of Defense, Memorandum of Understanding with the Russian Ministry of Defence on flight safety over Syria, 20 October 2015).

The clauses are kept by means that are the services' own. A standing line between the two staffs, used daily. A written protocol between two navies on how their ships and aircraft meet, reviewed each year by the two navies alone (Agreement on the Prevention of Incidents On and Over the High Seas, 1972; Winkler, Cold War at Sea, 2000). Liaison missions each side keeps on the other's ground, as the occupying armies did in Germany for forty years under agreements between their commanders (Huebner-Malinin Agreement, 1947). And the exchange of prisoners or of officers, through an intermediary both sides accept. Each is a place for the pact's work as Part VI gave it: the limit said, the warning before it, the breach repaired.

The costs above, all summed, shape the joint operation and set the program's end of term. The doctrine playbook holds it in real time, refreshed by every observation that returns from instruments. It feeds the program playbook below, and that playbook emits the instructions. It aborts on the same price, read on the general equation again. When the rally fades, the return it brought on the standing goal no longer covers the work of the operation: the days the forces spend, the labour behind the materiel the war consumes and the labour of holding what it ties, each priced at the bits it could have landed elsewhere. The sum turns positive. When the force raised against the enemy would have to be used on the governed, the act is seen, p(attribution) goes to one, and the scope condition strikes it. No goal on the sheet names coercion. The playbook discards it as a means on its own, as Part V said, and reads its use on the governed as a price it cannot pay.

The arithmetic is not the operation's. Every program in the portfolio runs on the same cost function, in the same unit, through the same factor k, with its own terms in the soldier's place: the pupil's hours, the commuter's, the user's, the voter's contact, the patient hospitalized. Each is priced per head, summed over its audience, and run while the sum stays negative and the resources allow it. A resource has bounds of its own: a reform takes its time, and a law changes slower than a playbook would. Those bounds sit in the playbook's models beside the prices. The operation example above only writes the sum where it is easiest to see.

Each of the five, taken alone, is a public policy with a public rationale: education, transit, investment, elections and defence are debated, funded and reported in the open. What none of them states is the framework behind them: the sheet of goals, the playbook that reads it and allocates the programs, and the circle from which both stem, its principles, its structure and its seats. Each program has a public author, a ministry, a planning authority, a fund, a party or a general staff, and none of these holds the sheet or the playbook, only the circle does. The parts are open. The composition is what brings understanding.