Self-criticism

There was a moment when this page carried what the essay did not say: the concessions that were missing and the defence that sustained the model. That has changed, and the change is the reason this page exists in another mode. The essay now accepts, in its own body, that relational geometry is not its invention, that the inversion of recipient had already been carried out by others, that the seven themes are a prior imposition upon the data, that the metric needed to be declared and that the experiment’s data are illustrative. The weaknesses recognized are no longer outsourced to a footnote.

What remains for this page is harder and more useful: to record what continues without solution after all the concessions made.

It is not a list of rhetorical humilities. It is a liability.

During part of the construction of this project, the central allegation was an inversion of recipient: the calculation of political space would always have been performed upon the citizen, by third parties, and this model would hand it to him. That allegation is false, and section V of the essay now withdraws it. It is worth recording here how it fell, because the manner of the fall matters more than the fall itself.

There is a consolidated genre of instrument that does exactly that. The so-called voting advice applications inform the user of his proximity to the parties’ positions, indicating which of them are closest to his preferences. The theoretical basis is Downs’s proximity model, of 1957. And most of them allow the voter 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.

These are not marginal experiments. The German Wahl-O-Mat had more than twenty-one million users in the 2021 federal election; Stemwijzer and Kieskompas together reached some six million eight hundred and fifty thousand users in the week before the 2017 Dutch election alone. There is an academic literature devoted to measuring their effects, and there is work published in 2012 that used data from that type of application to investigate the dimensionality of the political space in four countries, Brazil among them.

The consequence is direct and should not be softened: the inversion of recipient is not original. The citizen who reports positions, declares weights and receives the result is a practice established for decades, at a scale of tens of millions of people. Any reader familiar with the Dutch, German, Austrian or Swiss electoral experience recognizes this immediately.

What is left after that

Less is left than the project supposed it had, and it is honest to size with precision what survives.

A difference of representation survives. Kieskompas plots voter and parties in a two-dimensional space with predefined axes; the Swiss smartvote analyses eight dimensions and presents the result in a spider-web diagram; most of them return an ordered list of compatibility. None of those formats shows the field as a relational space seen from the observer’s position, in which the distances between the parties themselves also rearrange according to the weights he declared. A ranking informs the user whom he is closest to; it does not inform him how the field is structured when seen from where he stands. These are different objects, and that difference is real.

A difference of purpose survives. Those applications are electoral instruments, tied to a ballot, whose implicit recommendation is to vote for the closest party. The Political Constellation does not claim to recommend a vote: it claims to expose the structure of the field and the dependence of that structure on whoever observes, as a response to the algorithmic mediation of political perception. It is a distinct normative target.

The essay was corrected in consequence of this, and the correction was not confined to a paragraph of caveat: it altered the opening, the narrative of the genesis, the section on lineage, the formal definition, the civic function, the delimitation of what the model does not claim to be, and the closing. The resulting thesis is smaller and verifiable — the political field as a configuration seen from the observer’s position, and not the handing back of the instrument, which had already been done. To record the fall of a central allegation is uncomfortable; to publish it is the only way for the rest of the text to go on deserving credit.

The essay rests its civic function on a robust finding: once perceptual bias is corrected, a good part of the electorate turns out to be more extreme than was thought, and a considerable fraction of the self-declared moderates is classified as extreme by voting pattern — illusory moderation. The combined reading is elegant: the citizen perceives reasonably well where the others are and errs about where he himself is.

The logical problem is that this finding demonstrates the existence of the blind spot. It does not demonstrate that handing over an instrument of self-placement corrects it. These are different propositions, and the essay slides from one to the other without proving the passage.

Worse: there is reason to suspect that the instrument may aggravate the problem.

If rationalization bias consists in pushing away from oneself what one rejects and drawing close what one prefers, then an instrument in which the user himself declares his positions and his weights offers that bias a more precise channel, not an obstacle. The result may be a reproduction of the error with the appearance of measurement — and with greater confidence, because it now comes accompanied by numbers and by a three-dimensional image. A sophisticated error is harder to abandon than a guess.

What the literature offers in defence, and what it does not offer

In the model’s favour, the research on those electoral applications supplies precisely the evidence of treatment that was missing. It has been observed that individuals 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 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 list may lead to an effective change of vote — which is, in practical terms, the confrontation with illusory moderation taking place.

That evidence has been incorporated into the essay’s civic function, together with the finding that limits it — that instruments of this kind are used disproportionately by people who are younger, better educated and already interested in politics. The caveat that remains is of another order: that evidence was produced upon instruments with validated data, tested questions and positions collected from the parties themselves. It does not transfer automatically to a prototype that operates with approximations. So long as this model’s data are illustrative, the favourable evidence belongs to other instruments, not to this one.

The essay’s formal definition recognizes that seven was the number that seemed to cover the Brazilian public debate at the moment the prototype was built, without theoretical justification of its own, and as arbitrary as Nolan’s choice of two. It also recognizes that authority over which dimensions exist ought to belong to whoever observes, and that the prototype does not yet permit this.

What the essay does not examine is whether the promised solution really is a solution. If the observer chooses the dimensions, chooses the weights and occupies the origin, then practically everything in the space comes to be supplied by him. An instrument in which the user defines what to measure, how much each thing weighs and from where one looks runs the risk of returning, with increasing precision, what the user already thought before he began.

Total freedom over the construction of the space and the capacity to inform something new are objectives in tension, and the essay treats the first as though it were always a virtue.

There are possible ways out — offering a set of themes with a declared justification and permitting additions, or displaying simultaneously the sky chosen by the user and the equally weighted sky, so that the difference between the two becomes information. None is implemented. The problem is recorded as a problem, not as a stage.

All the other weaknesses on this page are theoretical: they concern what the model alleges. This one is operational, and it is the only one capable of directly harming a concrete person.

The party position data of the experiment are approximations drawn from literature, not validated measurement. The methodological note says so, and will go on saying so. But there is an uncomfortable asymmetry between the strength of the output and the fragility of the input: whoever uses the experiment receives an ordering of proximity with the appearance of a result, and the warning about the nature of the data arrives in text, outside the experience. A citizen who leaves the screen believing that a given party is the closest to him was informed by approximations, however many notices there may be around it.

Until that changes, the model’s civic promise exceeds the model’s civic delivery — and it is precisely the civic promise that the essay elected as its central justification.

This is the point at which the project is most exposed, not through an error of reasoning, but through disproportion between what it announces and what it can already do.

It is also placed on record 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 upon partial data, and should not circulate as an empirical conclusion under any circumstances.

The essay records, in the section on the Star Sextant, that the perceptual correction methods of this lineage exhibit a bias in the direction of unidimensional results, and that this bias is accentuated for respondents of low political information. It treats this as a pending task, to be resolved with anchoring vignettes and the explicit modelling of rationalization bias.

It may be more than a pending item. If dimensional flattening worsens precisely where political information is scarce, and if the target public of an open civic instrument is composed for the most part of people in that condition, then the instrument tends to return a flatter sky precisely to those to whom it meant to show that the sky is not a line. The promise and the mechanism point in opposite directions in the stratum of the population that matters most.

It is not known, today, whether the available corrections are sufficient to overcome that effect in an instrument of seven dimensions answered in a few minutes by someone with no prior interest in the subject. Until that is known, the allegation that the model reveals dimensional plurality to whoever saw only a line remains a hypothesis.

There is a recurrent finding in the research on analogous instruments that strikes at the heart of the civic justification: they are used disproportionately by people who are younger, have more formal education and have a higher prior interest in politics.

The perceptual emancipation promised is distributed, in practice, among the already emancipated.

If that pattern repeats here — and there is no reason to suppose it will not —, then the instrument reaches preferentially whoever already examines his own positions, and does not reach whoever accepts closed packages without reviewing them. This does not invalidate the project, but it does invalidate the rhetoric of universal reach, and it obliges the civic function to be formulated in more modest terms than the essay formulates it today.

The essay anchors its urgency in a broad hypothesis, developed in the source texts of the Algorithmic Zeitgeist: the fragmentation of the spirit of the age into algorithmic micro-hegemonies, the crisis of the public sphere, the algorithm as epistemological operator. It is a frame of civilizational scale.

What is delivered, today, is a space of seven dimensions with approximate data, navigable in the browser. The disproportion between the frame and the artefact is real, and attentive readers will feel it before they are able to name it. Announcing intention before possessing capacity has a cost, and the cost is this: the frame lends gravity to the artefact, and the artefact, if examined on its own, does not sustain it.

The honest alternative is not to abandon the frame — the hypothesis is the reason the project exists and is not incidental. It is to make explicit that the Political Constellation is a partial and initial response to a problem much larger than itself, and that the distance between the response and the problem is today considerable.

The relational tradition that the essay now recognizes as its own carries a consistent criticism: it is held that it privileges the relations between positions at the expense of the substance of those positions. The criticism applies with force to this model.

Knowing that one stands forty units of distance from a party on foreign policy does not inform one of what the divergence consists in, nor of what to do with it. Distance is a magnitude without content, and an instrument that returns only distances may produce the sensation of understanding without the understanding — which, in a project whose justification is autonomy of judgement, is a risk of the first order.

A user who discovers that he is far from every party has not gained autonomy; he has gained a measure of his own isolation.

Nothing in the current model resolves this. A future version would need to attach, to each distance, the content of the divergence that produces it — and that is a requirement of data and of curation, not of geometry.

Not every weakness pointed out here invalidates the project, and it is worth recording what withstands the examination.

The distinction between a vertex imposed from outside and a vertex anchored by whoever observes withstands it. A traditional cartesian axis has its origin defined before any datum enters; here, each theme is a vertex without a centre until a concrete person takes a position. That distinction is not enough to claim novelty, but it correctly describes what the model does.

The relational representation withstands it, having become the central allegation after the revision. To show the political field as a configuration that reorganizes itself according to the observer — including the distances between the parties themselves, and not merely between the observer and each of them — is different from returning a degree of compatibility, a ranking or a projection onto predefined axes. That difference is occupied by no known instrument, and it is small enough to be verified, which makes it preferable to the larger allegation that preceded it.

Finally, the frame withstands it. The hypothesis of the Algorithmic Zeitgeist does not depend on this model in order to stand, and it gives the model a purpose that the analogous electoral instruments do not have: not to recommend a vote, but to expose the dependence of political perception upon whoever organizes it. That purpose remains legitimate even if the current implementation is far from it.

It is possible that this set will not convince every critic that the model distinguishes itself sufficiently from its predecessors. But it is the most honest allegation that this project can make about itself today — and an honest allegation, however modest, is worth more than an impressive allegation that does not withstand the first careful examination.

Gathered here so that nothing on this page functions as an advance absolution.

First. Replace the illustrative data with verifiable sources, and make the nature of the data visible inside the experience itself, not only in a footnote. It is the only weakness capable of directly misinforming a concrete person, and for that reason it precedes all the others.

Second. Test whether the instrument corrects or reinforces rationalization bias, which requires measurement before and after use and cannot be presumed from results obtained with other instruments.

Third. Verify empirically the dimensionality of the perceived political space in Brazil with the data collected here, instead of assuming it, following the route already opened by work that used data from proximity instruments for that purpose in four countries.

Fourth. Allow the observer to choose his dimensions, and resolve simultaneously the tension between freedom of construction and the capacity to inform — possibly by displaying, in parallel, the sky weighted by the user and the equally weighted sky.

Fifth. Attach content to the distances, so that the measure of divergence comes accompanied by the divergence that produces it.

Sixth. Confront the vulnerability to strategic manipulation: the more the instrument circulates, the more advantageous it becomes, for whoever is measured by it, to present himself so as to be measured well.

Have you found a weakness, an incoherence, a blind spot that this page has not recorded? To send it is a real contribution to the project — public criticism is part of the method, not a nuisance. Every serious observation is read and considered, and those that expose an error of fact go onto this page with the name of whoever pointed it out, if authorized.

Let it be recorded that the gravest item on this page was not found by an external critic: it appeared during a deliberate adversarial examination of the model itself, and it brought down what was, until then, the project’s central allegation. That is no cause for pride — it means only that the examination needs to continue, and more often than vanity would recommend.