The Political
Constellation

Most theoretical models are presented already finished, as though they had been born whole out of an isolated intuition. This text proposes the opposite: to record the very reasoning that produced it, because the way a hypothesis is gestated explains as much as the hypothesis itself — and because the route, in this case, taught more than the destination.

The starting point was as elementary as possible, of the kind usually considered too naive to be formulated out loud: why is politics described by a line with two poles? The answer is historical, not theoretical. The physical arrangement of the deputies in the French National Assembly of 1789 — monarchists to the right of the chair, reformists to the left — turned an accident of seating into a permanent metaphor.

Discovering this produces a specific discomfort, and it is worth naming, because it was that discomfort which pushed everything that came afterwards. It is not the discomfort of knowing that a model is imperfect — every model is. It is the discomfort of realizing that one had been using, in order to think about one’s own public life, a ruler whose graduation nobody chose, and whose origin is an eighteenth-century meeting room.

The route that followed passed through Eysenck, through the horseshoe, through the two-axis diagrams, and arrived at a turn that seemed entirely new — until it met, further on, a set of authors who had already travelled much of the same road, with greater rigour and fifty years’ head start. That encounter is recorded in section V, without mitigation, because it was the most instructive part of the process. To discover that the intuition was not original, but was correct, is an uncommon form of good news: it means the problem had been rightly identified, and that the instruments for attacking it already existed, waiting for a use nobody had given them.

There was, afterwards, a second discovery, harder than the first, and it is recorded in section XI. Not only did the instruments already exist: so did the use. Instruments that hand the citizen the calculation of his proximity to the parties have been operating for decades, at a scale of tens of millions of people. The question then had to be asked a third time — no longer “how many axes”, nor “who receives the result”, but: what exactly do those instruments show, and what do they not show? The answer to that last question is what remains this model’s own, and it is deliberately narrow. (self-criticism)

Before the geometry, the reason for being. This model was not born out of an interest in spatial formalisms; it was born out of a prior hypothesis about what happened to the spirit of the age, and it is from that hypothesis that it draws all its urgency.

For centuries, the Zeitgeist seemed almost invisible. It shaped behaviours, values, beliefs and collective perceptions without most people clearly noticing its operation: one was born inside a cultural atmosphere and absorbed it as natural. Culture operated silently, and at slow speed. The formation of consensus depended on stable institutions — schools, universities, churches, newspapers, radio, television. The twentieth century was perhaps the height of that model: a few broadcasters transmitted symbols and narratives to millions simultaneously, and there was a dominant cultural narrative strong enough to organize the collective perception of reality.

Digital platforms structurally altered that dynamic. What was vertical became horizontal; what was slow became instantaneous; what was relatively centralized became fragmented, algorithmic and permanently mutable. Today there is not merely one dominant Zeitgeist: there are multiple micro-Zeitgeists coexisting, each bubble developing its own language, its own references, its own values, its own enemies and its own forms of moral and political perception.

The algorithmic micro-hegemonies

This phenomenon has a name, and it derives directly from updating the Gramscian theory of cultural hegemony for the techno-digital environment. Where Gramsci described power sustained by the capacity to make certain values be perceived as natural, the platforms produce not a relatively unified hegemony, but multiple competing poles of influence mediated by algorithms. Each digital community comes to operate inside a partially closed semantic universe, reinforced by recommendation systems that intensify identity belonging and ideological confirmation.

The decisive consequence is epistemological, not merely communicational: the algorithm ceases to be a mere technical intermediary and takes on an epistemological function. It organizes what may be seen, discussed, legitimated and emotionally intensified. There is no longer a single, widely shared spirit of the age, but multiple competing regimes of social perception, each sustained by specific architectures of visibility.

If each micro-hegemony is a competing regime of perception, then each micro-hegemony is an observer — and each observer draws its own lines between the same points of light.

It is here that the prior hypothesis meets the need for a formal structure. Each observer produces distinct constellations out of a sky of facts that is only partially shared. The Political Constellation is the attempt to give calculable form to that statement.

Why this is a moral question, and not merely a descriptive one

Habermas held that modern democracies depend on shared spaces of rational debate oriented by argumentation. Algorithmic logic strains that model directly, because it privileges emotional intensity, speed and engagement, frequently at the expense of deliberation — and the result is a structural crisis of public rationality, in which society loses the shared epistemological references capable of sustaining minimal consensus.

It is worth being precise about the mechanism, because precision here is what separates an argument from an accusation. High-intensity emotions — indignation, fear, belonging, anxiety — receive algorithmic priority because they increase retention and sharing. The technical architecture of the platforms therefore structurally favours polarization, narrative simplification and affective radicalization. The point does not depend on attributing intention to anyone: it depends only on recognizing what optimization for engagement optimizes. Grouping by emotional affinity is not the same as grouping by programmatic proximity, and it is the first operation the systems perform, not the second.

From this follows the concrete consequence for the ordinary citizen: he comes to believe that there are only two camps, each internally cohesive, and that his task is to choose a closed package of positions — not to assess, theme by theme, what actually matters to him. The ruler of 1789 survives not by theoretical virtue, but because it is the most compact way of organizing affect at scale.

This is a frame of civilizational scale, and what is delivered today is a prototype navigable in the browser. The disproportion between the two things is real and is examined in detail in the self-criticism.

The first relevant attempt to cross the left-right axis with a second dimension did not come from Faye, but from Hans Eysenck, almost a decade earlier. In The Psychology of Politics, of 1954, Eysenck proposed a second axis — which he called tough-minded against tender-minded — and, from factor analysis of questionnaires, argued that fascists and communists scored similarly on mental toughness and authoritarian submission, differing from centrists.

Only later, at the beginning of the 1970s, did the French philosopher Jean-Pierre Faye popularize the visual metaphor of the horseshoe, in Langages totalitaires: instead of moving indefinitely apart, the extremes draw close at the tips. He should not be confused with Guillaume Faye, an author of another political tradition, to whom he bears no relation. And even that attribution is disputed by historians of the idea: the horseshoe scheme, Hufeisen in German, was already circulating in German political discourse from the 1930s onwards, tied to diagrams of the Weimar Republic period. Faye’s text is reporting a pre-existing scheme rather than inventing a new one.

From Eysenck to Faye, what recurs is a real yet partial correction: both still assume a line, or an axis, as the underlying structure, merely crossed or bent. The underlying problem — a single axis, imposed from outside — remains intact. The curvature was changed, or a dimension was added; not the logic.

Let it be recorded, in the horseshoe’s favour, that it did not remain merely a metaphor: there are models today that project positions onto the unit circle precisely in order to capture the behavioural convergence between the extremes and to correct the distortions of projecting political behaviour into a linear space. Faye’s intuition gained formalization. What it did not gain was freedom from the axis.

The libertarian activist David Nolan published his two-axis diagram — economic freedom crossed with personal freedom — in 1971, still in the same decade as Faye. Thirty years later, in 2001, the Political Compass appeared, crossing an economic axis with an authoritarian-libertarian axis in the format of an interactive site; its creation is attributed to the journalist Wayne Brittenden, although the site itself identifies no authorship. Between one and the other lie almost three decades, which makes it inadvisable to cite them side by side as though they were contemporaries.

The natural question is the same for both: why stop at two dimensions? Increasing the number of axes does not solve the original cartesian problem; it merely disguises it in more dimensions, because someone continues deciding, from outside, what each axis means.

And there is a technical limit that proves more severe than intuition suggests. In spaces of many dimensions, the distances between points tend to concentrate: the ratio between the nearest and the farthest neighbour approaches one. The problem is not that each point becomes isolated from the others; it is that all of them come to lie at approximately the same distance from all the others, and the very notion of proximity loses its power of discrimination. It is the curse of dimensionality in the form most unfavourable to any thematic model.

There is a point beyond which more dimensions produce a sky in which nothing is near anything.

The warning applies against this model as much as against the others, and is recorded as a limit on growth: adding themes indefinitely does not increase the information available, it degrades it.

The conceptual turn happened when the question ceased to be “how many axes do we need” and became another: what if we were not on the plane of a chart, but in space — where there is no single sun pulling every body towards a centre, only bodies among themselves, drawing closer or farther apart by affinity of thought?

The formulation, when it appeared, had the quality of ideas that seem obvious once stated: not a political cube instead of a line, but a purely relational space, in which the only input is the proximity between positions, and in which the geometry — including the existence or not of axes — is a consequence of the data, not a prior imposition upon them.

The discovery that reorganized the project

It was in looking for whoever might already have formulated something of the sort that the project found its most instructive moment. The answer is: practically everyone, and long ago.

The class of techniques whose aim is to locate objects in a low-dimensional space in such a way that the distances between the points correspond to observed dissimilarities was formulated by Warren Torgerson in 1952, with a precursor in the Young and Householder procedure of 1938. The input is only a matrix of proximities; geometry is output, not input. In 1962, Roger Shepard treated the problem with an unknown distance function — that is, without stipulating the metric of the space in advance. And in 1964 Joseph Kruskal converted the intuition into a method, defining an explicit loss function, the stress, which turned the question “how many dimensions?” from a doctrinal matter into an empirical one: stop where the stress ceases to fall significantly.

But the finding that matters most is from 1970. In that year, J. Douglas Carroll and Jih-Jie Chang published a model of individual differences in which a common space of stimuli, shared by all, is postulated, and in which each individual weights each of the dimensions of that space differently. The individual weights systematically distort the common space and produce, for each person, what the literature calls a private space. The distances in the common space are weighted euclidean; in the private spaces, simple euclidean.

That is, with exactitude, the mathematical architecture of the Political Constellation. Not an analogy: the same class of model.

A set of parties in a common space of thematic dimensions; a set of weights per observer; a private sky as the result. Recognizing this costs the claim of formal originality, and the trade is advantageous on three fronts: a declared metric, a regime of falsifiability, and more than fifty years of literature on the properties and the limits of the model.

Above all, it makes it possible to state with precision what remains its own — and what remains is an inversion narrow enough to be verified. In the 1970 model, the weights are estimated by a researcher, from judgements collected from the subject, in order to explain why different people perceive the same set of stimuli differently; the private space is a finding about the subject, and it goes to the researcher. Here, the weights are declared by the subject himself, and the private space is the product delivered to him. The operation is the same. The direction is inverted.

The applied lineage: the instruments that already hand the calculation to the citizen

There is a second lineage, and it is not methodological: it is one of application. Its theoretical foundation is the proximity model formulated by Anthony Downs in 1957, according to which the voter tends to prefer the alternative closest to his own positions. On that foundation there was built, from the nineteen-nineties onwards, a family of instruments that informs the user of his proximity to the parties’ positions, indicating which of them are closest to his declared preferences.

It must be said clearly what this means for the present project. Most of those instruments allow the user to increase the weight of the themes he considers especially important and to reduce that of the ones to which he is indifferent, with that weighting incorporated into the calculation of the distance. The mechanism is therefore the same as the one described here: positions declared by the observer himself, weights declared by the observer himself, weighted distance as the result, delivered to whoever answered. And the scale is not experimental: a single one of those instruments, in Germany, was used by more than twenty-one million people in a federal election; in the Netherlands, two of them together reached some six million eight hundred and fifty thousand users in the week before a national ballot alone.

The inversion of recipient, taken in isolation, is therefore not a contribution of this model. Recognizing this is an elementary obligation, and what is left after the recognition is examined in section XI, where the criticisms that this family of instruments receives, and that apply to this one in full, are also recorded. (self-criticism)

The observer who distorts the map

A second encounter confirmed that the central intuition was not merely defensible, but measurable. In an article of 1977, John Aldrich and Richard McKelvey introduced a method for correcting what is called differential item functioning: the fact that respondents perceive and use the same ideological scale in fundamentally different ways, producing divergent placements of the same stimuli. There is a sub-case that is, word for word, one of the properties of this model, and it is called rationalization bias: the tendency to push away from oneself the positions of the parties one rejects and to draw close to oneself those of the parties one prefers.

The property that this essay calls the observer who draws the lines is therefore not poetic licence imported from astronomy. It is a phenomenon that has been measured for almost fifty years, and the hypothesis of the micro-hegemonies gains from this something more valuable than a metaphor: it gains a method by which it could be tested.

The relational tradition

The refusal of a hierarchy external to the phenomenon is not, either, a recent invention. Relationality is the centre of Pierre Bourdieu’s sociology, to the point that he preferred to speak not of a theory, but of a system of relational concepts, and to maintain that the real is relational; distinction, for him, is nothing other than difference, gap, distinctive trait — a property that exists only in and through its relation with other properties.

What matters is the method. In Distinction, of 1979, Bourdieu applied multiple correspondence analysis to data on income, occupation and cultural practices, positioning individuals, occupations and activities on a plane, and only then interpreted the axes he had found: the first dimension as total volume of capital, the second as the contrast between economic capital and cultural capital. Note the logical order: the axes were not stipulated and then filled with data; they were derived from the data and then interpreted. It is the exact inverse of the political compass, and it was done with data forty-seven years before this essay.

Honesty obliges one to record the criticism that the lineage carries, because it will apply to this model as well: it is held that Bourdieu privileges the relations between positions at the expense of the substance of those positions. A model that reports only distance runs the risk of ending up mute about the content of what separates. It is a real vulnerability, and there is no reason to conceal that it is shared. (self-criticism)

On Benjamin, Adorno and Habermas

There remains a coincidence of vocabulary that deserves to be named rather than hidden. Walter Benjamin proposed, in a work completed in 1925 and published in 1928, that ideas are to objects as constellations are to stars; Theodor Adorno carried the image further as an explicit method in Negative Dialectics, arranging concepts around an object without forcing a classificatory synthesis; Jürgen Habermas used the term in a political sense in Die postnationale Konstellation, of 1998.

The mention exists for the sake of disambiguation, and not of inheritance. The Frankfurt Konstellation is an epistemological metaphor about how abstract concepts relate to one another in philosophical prose; this model calculates distances between real positions around a situated observer. The kinship of spirit — refusing a hierarchy imposed from outside — is genuine, but the methodological kinship is with Torgerson, Carroll, Aldrich and Bourdieu. The one that carried out the refusal with data was relational sociology; German philosophy offered the image, and it is fair to say only that.

There is a serious objection to face, and it does not come from philosophy: it comes from the data. When the positions of legislators are estimated from roll-call votes, without imposing any axis at all, the left-right axis reappears on its own. In the Congress of the United States, it has been shown that the votes can be explained by no more than two dimensions across almost the whole history of the country, with a recent collapse into a single dimension. The axis does not survive merely by the inertia of 1789: in certain contexts, it emerges from the data when nobody imposes it.

The answer is not to deny the finding, but to observe what exactly it describes — because what it describes is precisely the world whose dissolution the Algorithmic Zeitgeist documents.

First, those findings measure the voting behaviour of parliamentary elites operating under party discipline and coalition calculation. That is not proximity of content as perceived by citizens, and the distinction is not rhetorical: a legislator may vote with the government out of coalition arithmetic and diverge from it on most of the themes that matter to those who elected him. The line captures the vote; it does not capture what the voter is trying to find out. Effective alignment in the legislature and the position perceived by the electorate are distinct quantities, and it is in the interval between the two that this model operates.

Second, and more importantly: unidimensionality is the statistical signature of a regime of centralized symbolic mediation. It describes societies in which few broadcasters organized collective perception, in which there existed a dominant cultural narrative strong enough to align positions along a single continuum, and in which consensus settled slowly through stable institutions. The more centralized the mediation, the fewer dimensions are needed to describe the result — and that is not coincidence, it is consequence.

What the hypothesis of the Algorithmic Zeitgeist maintains is that this regime has ended. Cultural hegemony has atomized into competing micro-hegemonies; the public sphere has fragmented into micro-spheres without stable common references; the unifying legitimating narratives have collapsed into multiple algorithmically mediated regimes of truth. An instrument calibrated to measure the earlier world will go on working well where its conditions remain — inside a chamber, under party discipline, with a controlled agenda — and will go on being blind to whatever has organized itself outside.

The single axis is not this model’s adversary. It is its baseline.

The question this essay asks is not whether the line ever described anything well; it is what remains of it once the mediation that produced it has disappeared, and once whoever observes the political sky is no longer a relatively unified public, but someone inside one regime of perception among many.

There is, moreover, an empirical path already open for investigating this, and it does not run through roll-call votes. Data produced by proximity instruments answered by citizens were used, in work of 2012, to explore the dimensionality of the policy space in four countries, Brazil among them. That is to say: the question of how many dimensions describe the perceived political space can be attacked with the same kind of data that this model collects — and it should be, before any claim about Brazilian dimensional plurality is made from here.

It is best not to convert this into an empirical claim larger than the evidence permits. That the political space perceived today is irreducible to one dimension is a hypothesis, not a demonstrated result. The difference between this text and a metaphor is that the hypothesis, here, is formulated so as to be testable — and the following section declares with what instrument.

The concession, first

The most serious criticism this model receives is also the simplest to formulate: beneath the stellar metaphor, each party is positioned by seven numbers — economy, social values, institutions, foreign policy, public safety, environment and technology —, each one ranging from minus one hundred to plus one hundred. That is, technically, a cartesian system of seven dimensions. The model did not eliminate the axis; it multiplied it by seven.

The criticism is correct and should be accepted without mitigation. The choice of the seven themes is a prior imposition upon the data, of the same kind that Nolan made in 1971 when he elected two vertices. And the number has no theoretical justification of its own: seven was what seemed to cover the Brazilian public debate well at the moment the prototype was built — a pragmatic choice, as arbitrary as Nolan’s choice of two. (self-criticism)

The precise formulation, therefore, is this: what this model does not impose from outside are not the variables. They are the origin and the weighting. The dimensions are declared by whoever builds the model, and that is an admitted limitation. What belongs to the observer is where the zero sits and how much each theme weighs. A traditional cartesian axis has its origin defined beforehand — the zero of the ruler is already there before any datum enters. Here, each theme is a vertex without a centre until a concrete person takes a position; only then does it become a real coordinate, and only for that observer, at that moment.

The metric

The thirty parties registered with the Superior Electoral Court occupy a common space of seven dimensions. The observer occupies a position in that same space, determined by his answers to the Star Sextant, and declares a weight for each theme according to the importance he attributes to it. The distance between the observer and each party is a weighted euclidean distance:

where o is the observer’s position, p the party’s position and w the weight the observer assigned to that theme. When all the weights are equal, the expression reduces to the simple euclidean distance: the equally weighted sky is only a particular case of each person’s sky. The normalization of the weights and the treatment of themes with zero weight form part of the experiment’s documentation, so that the calculation is reproducible by third parties.

Declaring the metric has a consequence that matters more than precision: it makes the model falsifiable. It is possible to verify whether the calculated distances predict declared voting intention, or thematic agreement measured by other instruments, better or worse than self-placement on a left-right scale. So long as distance remains a metaphor, there is no way for the model to be wrong — and a model that cannot be wrong informs nothing.

It should be recorded here, and not in a note, that weighting distances by declared importance is not an unprecedented procedure: it is current practice in the proximity instruments described in section V, in which marking a theme as important usually doubles its effect in the calculation, and marking it as indifferent usually halves it. What differs in this model is not the weighting, it is the output. Those instruments return a degree of compatibility — a number, a ranking, or a projection onto predefined axes. Here, the output is a configuration: besides the distance between the observer and each party, the distances of the parties among themselves are altered too, because all of them are recalculated under the same weighting. The user does not merely receive the answer as to whom he is closest; he receives the shape the field assumes when measured by his own criteria.

That is a difference of representation, and it is smaller than the difference this essay once supposed it had. It is also verifiable, which makes it preferable to a larger and vaguer allegation.

The nature of the data

The party position data used in the experiment are illustrative, approximated from political science literature on the Brazilian party system. They do not constitute validated measurement, and the experiment should not be read as such. A production version would replace those approximations with verifiable sources: the Manifesto Project’s Rile index, expert surveys, or recent roll-call votes disaggregated by theme.

Two consequences are placed on record. The first is that the finding that no registered Brazilian party crosses the far-right threshold under the symmetrical formula adopted is a real result of the calculation, but one supported by partial approximations, and it should not circulate as an empirical conclusion. The second is that the coverage of legislative votes by theme is still fragmentary, which limits any claim about a party’s position on a specific theme. (self-criticism)

The direction of development

If authority over the origin of the space belongs to whoever observes, authority over which dimensions exist in that space ought, in principle, to belong to him as well. The current prototype does not permit this: the list of seven themes is inherited, not chosen. That is the model’s remaining incoherence, and the way out is not to find the right number of themes — it is to allow each observer to assemble his own configuration of vertices. It is recorded as a limitation of implementation and a direction of work, not as a closed decision. (self-criticism)

With the dimensions and the metric described, five properties distinguish the model both from the cartesian line and from the horseshoe and the fixed-axis diagrams.

Temporal snapshot. The model does not describe a permanent structure, but a cross-section valid for the instant in which it is observed. Redo the observation, with other weights or other data, and the constellation reconfigures itself.

Apparent distance. The proximity observed between two positions is never absolute — like two neighbouring stars in the night sky separated by immense real distances. Two positions may appear close only because the angle of observation does not reveal the dimension on which they diverge.

The observer who draws the lines. The figure of the constellation does not exist objectively between the points: it is constructed by whoever observes. This property requires, however, a distinction at three levels, without which the model loses the right to criticize anyone else’s observation — and that is the subject of section IX.

Unequal magnitude. Not every point attracts with the same force. Reach, engagement and emotional intensity work like the brightness of a star — and brightness is not proximity: the most visible position in the political sky is not, for that reason, the closest to whoever sees it.

Overlapping belonging. One and the same position may belong to multiple constellations simultaneously, one per contested theme. What on the single line appears as contradiction is, here, a normal condition.

A model may be conceptually elegant and still fail to answer the most basic question: what is it good for, in practice, to someone who is neither a political scientist nor a software developer? The answer begins where the Algorithmic Zeitgeist left off.

Three levels, so as not to lose the right to criticize

If every constellation is the work of an observer, then the constellation drawn by a platform is as legitimate as anyone else’s, and criticism of the algorithm collapses by internal contradiction. To speak of real proximity of position and, at the same time, of the objective non-existence of the constellation requires separating three levels.

There is, first, similarity of content between positions: two programmes agree or diverge on each theme, and that is verifiable through documents, votes and declarations, independently of who observes. There is, second, apparent proximity, which depends on the point of observation and on the weighting — and that varies legitimately from person to person, because different people attribute different importance to the same themes. There is, third, the question of who defines the point of observation.

The criticism of the algorithmic map is not a criticism of accuracy. It is a criticism of autonomy.

The problem is not that the platform sees wrongly: it is that the point of view from which it draws was not chosen by whoever receives the drawing, and the weights it applies are not the ones the citizen would declare if he were asked. Weighting by engagement is legitimate for the purposes of whoever applies it. It simply is not the citizen’s — and he was not consulted.

The datum that justifies the instrument

There is empirical reason, and not merely normative reason, for wanting to give the instrument back to those who are measured by it. It comes from the lineage of perceptual correction begun in 1977 and consolidated in a Bayesian implementation published in 2015.

Differential item functioning biases self-placements on an ideological scale in such a way as to underestimate the real polarization of the electorate. Once the bias is corrected, forty-four per cent of respondents appear as more extreme than either of the two presidential candidates in contention; and among the self-declared moderates, thirty-four per cent are classified as extreme on the basis of voting patterns — a phenomenon described in the literature as illusory moderation. At the same time, the perceptions that citizens have of the position of others, once corrected, correlate strongly with effective behaviour.

The citizen perceives reasonably well where the others are. He errs systematically about where he himself is. Self-perception is the blind spot.

And that is precisely what no existing instrument gives back to him. The concrete example is worth more than the statistic. In 2012, in the United States, almost a third of the respondents who voted for Barack Obama classified both him and themselves as moderate or slightly liberal, while some sixty per cent of Mitt Romney’s voters placed Obama at the furthest left position possible on the scale. The same object, two different skies — and, in both, a self-placement more comfortable than the vote could sustain.

The evidence that the instrument works, and the evidence that it does not reach those it should

The finding of illusory moderation demonstrates the existence of the blind spot. It does not demonstrate, by itself, that handing over an instrument of self-placement corrects it — these are distinct propositions, and the passage between them needs evidence of its own. That evidence exists, and it comes from the proximity instruments already mentioned. (self-criticism)

It has been observed that users with medium or low levels of formal education gain an expanded awareness of the parties’ positions and of the differences between parties, and that this learning effect raises the probability of mobilization, of reconsidering a choice or of reinforcing a decision already taken. It has also been observed that the irritation provoked when the preferred party does not appear at the top of the result may lead to an effective change of vote. That is, in concrete terms, the confrontation with one’s political self-image taking place — and it is the best available evidence that instruments of this kind produce something more than civic entertainment.

The same literature, however, imposes a correction of expectation that this essay cannot omit: such instruments are used disproportionately by people who are younger, have more formal education and have a higher prior interest in politics. If the pattern repeats here, the instrument will reach preferentially those who already examine their own positions, and not those who accept closed packages without reviewing them. The civic function remains legitimate, but its real reach is narrower than the generous formulation suggests, and it is honest to say so before someone else does. (self-criticism)

The two functions

The first is diagnostic. By dissolving the closed package into seven themes weighted by whoever observes, the model reveals what the single line conceals: two people may consider themselves rivals and be close on four of the seven themes; two people on the same side may diverge drastically on the theme that matters most to one of them. And, above all, a citizen may discover that the party closest to his weighted positions is not the one whose label he carries. That is real information, not rhetoric — and today it exists only in fragmented form, inside systems that calculate it about him, never visible to the person they observe.

The second is cognitive protection. If the constellation that someone sees is drawn by whoever observes — and today, in practice, whoever observes at scale is the algorithm —, then knowing this is already a defence. To recognize that the proposition according to which left and right are opposed and internally uniform blocs is a line drawn by a specific observer, under a specific weighting, is the first step towards ceasing to treat it as a fact of nature. And the finding of illusory moderation shows that political self-image is constructed — which means that it can be examined.

It is best not to promise more than that. No model gives back autonomy of judgement to someone who does not wish to exercise it, and the claim that this one gives back anything at all is a normative wager, not a proven result. What is offered is one’s own perspective — the position from which each person observes the political sky — to whoever already intends to decide theme by theme, instead of accepting the closed package that the last social network visited decided to show.

A spatial model is only useful if whoever observes it can also locate himself within it. The solution is not to reintroduce axes, but to import from astronomy itself the tool that has solved that problem for millennia: navigation by guide stars.

The intended effect is to give the reader an almost physical experience of materializing inside the model itself, like an astronaut suspended in a field of stars.

This raises an immediate and legitimate objection: a guide star is a fixed point of reference, and the model has just refused fixed points of reference. The distinction exists and needs to be argued, not presupposed. A local frame of navigation and a global frame of classification differ in three respects. The first is plural: any sufficient set of stars will serve, and different navigators may use different sets to arrive at the same place. The second is instrumental: it serves to answer where am I in relation to this, not what this is. The third is revisable without cost: changing guide star does not reorganize the sky, only the manner of looking at it. A global axis of classification has none of those properties — it is unique, it is attributive, and changing it redefines every object it classified.

No navigator needs to understand the complete three-dimensional structure of the universe in order to find his bearings on the ocean; it is enough to recognize a few stars and to position oneself by relative proximity to them. It is that capacity for concrete self-location, while preserving the spatial fluidity of the model, that turns the Political Constellation from an elegant metaphor into an instrument with a real function for whoever reads it.

The difficulty the instrument must face

There is an objection that honesty obliges one to anticipate, and it is serious. The perceptual correction methods developed in this lineage exhibit a bias in the direction of unidimensional results, and that bias is accentuated precisely for respondents of low political information — who are, by definition, the most numerous public of any openly accessible civic instrument. A sextant intended for everyone tends to return, to those who most need it, a sky flatter than the real one. (self-criticism)

The literature offers routes: the explicit modelling of rationalization bias and the use of anchoring vignettes, which describe fictitious programmes in order to calibrate the scale without supposing that all respondents perceive the same actor identically. There is also an identification requirement relevant to the design of the questionnaire, and not merely technical: at least three external stimuli are needed for the correction to be identifiable. Incorporating this into the Sextant is a pending task, and is recorded as such.

The Political Constellation does not claim to replace the models of position estimation as a tool of scientific measurement. It does not compete with the estimation of ideal points from roll-call votes, nor with scaling from texts, nor with the Bayesian correction of perceptions in sample surveys. Those instruments are better than this one at what they do, and this model depends on them in order to obtain the data it needs.

Nor does it claim to be a voting advice application, and the distinction matters because the kinship is evident. Those instruments exist to answer a practical question — whom to vote for — and their implicit recommendation is to vote for the closest. This model does not recommend a vote, is not organized around an election and does not treat the shortest distance as advice. Its object is the structure of the field and the dependence of that structure on whoever observes. These are different purposes operating upon a similar mechanism, and confusing them would damage both.

This does not exempt this model from the criticisms that family receives, and all of them apply. Proximity instruments disregard the credibility, responsiveness and competence of the actors; they treat declared positions, and not delivered results, as what matters; and they are vulnerable to strategic manipulation, because a party that knows the instrument can declare positions calculated to improve its placement. The last vulnerability is serious here: the more a model of this kind circulates, the more advantageous it becomes, for whoever is measured by it, to present himself so as to be measured well. No solution to this is implemented. (self-criticism)

Nor does it claim to be a test of political identity, or to produce a verdict on who someone is. The model’s output is an ordered list of distances under a declared weighting, valid for the moment of the observation. Change the weighting and the order changes — and that is not a defect: it is what the instrument sets out to show.

What it offers is a correction of recipient. The need for fixed axes is not a requirement of political reality, and the first to demonstrate this, with data, was the tradition cited in section V. What nobody in that tradition did was to hand the result to the person who was being measured.

For more than seventy years, the relational space of politics has been calculated with growing competence. Psychometricians built the methods; political scientists estimated the positions of legislators; platforms estimated the positions of users; campaigns estimated the positions of voters. And, at some point in the nineteen-nineties, someone had the good idea of handing part of that calculation back to the person who was being measured — that is how the electoral proximity instruments were born, and this essay once supposed that this handing back had not yet occurred.

It did occur, and what remains to be done is more modest and more precise: those instruments hand back an answer, not a configuration. They say whom the citizen is closest to; they do not show the shape the field assumes when seen from where he stands, nor how the positions rearrange among themselves under the criteria he declared.

The question that was left without an instrument is not who represents me best, it is how the sky arranges itself when I am the point of observation.

At the same time, the spirit of the age has ceased to be a diffuse historical abstraction and has come to constitute a technically mediated process, continuously recalculated by systems that modulate perception, emotion and visibility on a planetary scale. Humanity has become, all at once, the subject, the observer and the operational object of its own cultural formation. It has never had so much access to the spectacle of itself being formed — and never so little access to the instrument that would allow it to measure where, exactly, it ended up.

The Political Constellation is the attempt to close that gap with modest means: a model from 1970, a declared metric, still imperfect data and a difference of representation that fits into a single sentence. It does not invent relational space, it does not invent weighting by importance, it does not invent handing the calculation to the citizen. It does one thing only, and it is the only one that had not yet been done: it shows the political field as a configuration, and it places at the origin of that configuration the person who is looking.

If the contemporary spirit of the age is continuously recalculated, then whoever is described by that calculation ought to be able to run it for himself. Not a fixed structure to be filled in with data, but the legitimate photograph of a sky that never stops moving, and that changes shape according to the observer who contemplates it — taken, this time, by the person who is inside it.