Methodology

There are 12 axes. Three of them form the cube (Economic, Authority, Identity), eight are on the default radar, and four more sit behind the 12-axis toggle. Each has an operational definition — what a government at each end actually does — because an axis labelled only with a direction can be filled in with whatever the reader already believes.

Factor analyses of political attitudes usually recover somewhere between two and four latent dimensions that explain most of the variance. Everything past that is not nothing — it is structure that the leading factors do not capture — but it is explanatory rather than fundamental, and it is often specific to a country or a decade.

So the 12 axes here are not a claim that political space really has 12 independent dimensions. The three cube axes carry most of the work. The radar’s other five are there because collapsing them into the first three loses distinctions that matter in argument, and the 4 optional axes (Ecology, Military, Class, Rationalism) are off by default because they correlate heavily with axes already shown.

The cube’s Z axis is the clearest compromise in the model. It bundles cultural traditionalism with national particularism, two things that usually move together and sometimes do not. The radar splits them into Identity and Nation. Both treatments are on screen at once, deliberately, so the seam is visible.

Distance between your position and a tradition is a weighted Euclidean distance over the axes currently active:

d(u, i) = √( Σ wₐ (uₐ − iₐ)² / Σ wₐ )

Dividing by the sum of weights is what makes distances comparable when you switch between 8 and 12 axes — without it, adding an axis raises every distance and the nearest label appears to move for a reason that is purely arithmetic. All weights are 1 in Phase 1; per-axis weighting (“borders matter more to me than tax”) is Phase 3.

Each axis spans [−1, +1], so the largest possible distance is 2. When the second-nearest tradition is within 0.12 of the nearest, the interface reports a tie instead of naming a winner. A nearest label is a neighbourhood you are standing in, not an identity assigned to you, and the arithmetic is often much less decisive than a single confident label suggests.

The observation behind horseshoe theory is real and narrow: movements at the far ends of the economic axis have repeatedly converged on the same means — emergency powers, control of the press, enemy lists, a party that is also the state. In this model that shows up as a genuine pattern on the Authority and Democracy axes.

What does not follow is that the ends converge, or that the movements are interchangeable. A collectivised economy and a corporatist national economy are not the same programme, and the people each targeted were not the same people. The cube is built to keep both facts visible at once: shared height on Authority, opposite positions on Economic. That is why the horseshoe overlay planned for Phase 4 will be off by default and will highlight means only.

The Overton window is the band of positions that can be advocated by a mainstream figure without ending their career. It is a fact about a particular country at a particular moment, not about the positions themselves — and it moves, sometimes quickly. Positions that sat inside it a generation ago now sit outside it, in both directions.

Nothing on this map is scored by whether it falls inside anyone’s window. The window belongs on top of the map as a movable band, which is Phase 4 work and needs country-and-year data this build does not bundle.

The origin of this map is a stipulated global midpoint, and that is an editorial choice with real consequences. A health policy that is unremarkable in the Netherlands is a left-wing position in the United States and a right-wing position in Cuba. The same person plotted against a US median and a Western European median lands in visibly different places.

Recentring the origin on a country or a year is Phase 4. Doing it properly needs published median positions — the Comparative Study of Electoral Systems, the Chapel Hill Expert Survey and the World Values Survey are the usual sources — and until those are bundled and cited, the honest thing is to say the origin is stipulated rather than to invent country medians.

Ask people to place themselves on a left–right scale and then compare the answer to their positions on specific policies, and the two come apart regularly. The same is true of politicians: self-described moderates vote with their party’s wing, and roll-call scaling methods routinely place legislators somewhere their own literature does not.

There are at least three reasons, and they pull in different directions. Labels are social identities as much as policy summaries, so people adopt the one their group uses. Scales are relative, so “moderate” means the middle of whatever the respondent considers the field. And roll-call records measure votes taken under party discipline on bills chosen by leadership, which is a different object from a person’s beliefs.

This tool measures neither. It measures where a set of stated positions sits relative to a set of stylised traditions — a third object again, and one worth being explicit about.

These placements are ones where the literature genuinely disagrees, and a single point misrepresents the disagreement. They are marked in the interface rather than smoothed over.

The complete table. 20 traditions × 12 axes, the numbers the cube, the radar and the ranking are all computed from.

Traditions are abstractions and readers reach for names, so the figure can also draw historical figures. Four rules govern the layer, and they are stricter than the ones for the traditions because naming real people raises the cost of being sloppy.

Programmes, not persons. Each placement reads what someone did in power — or, for the thinkers, argued for — and the record it reads is printed beside the coordinate in the tool itself, not only here. Three axes only: scoring these figures on all 12 would mean inventing 9 numbers per person to fill a radar with. A serving record is read closed: the layer once excluded anyone still in office, which kept out most of the figures readers arrive looking for. Instead, a serving figure is placed from a completed stretch of the record — a finished term, or the years already governed — is marked servingwherever it appears, and is capped at medium confidence however long that record runs, because the rest of it cannot be read yet. The filters beside the figure will take them back out. And disputes are declared, exactly as they are for the traditions.

Two classifications sit beside the coordinates, and they exist to let a reader cut sixty figures down to the handful they came for. Neither is derived from where anyone sits. Role records the chair they sat in — heads of state & government, legislators & officials, economists, philosophers & theorists, movement leaders, business & industry. Route records how they came to hold it: won a national vote, rose through a party or ministry, took power by coup, war or revolution, succeeded a family member, or never held office. Route is deliberately a fact about arriving in the office and not a verdict on what followed — several figures here reached power through a ballot and then dismantled the thing that put them there, and the basis beside each coordinate says so. Both are printed in the table below.

Each marker is drawn to whatever its figure is most recognisable by — a flat cap, a pair of round glasses, set hair — so a reader can find someone by silhouette before they can read a label. Nothing in that drawing is derived from where the figure sits: a silhouette assigned by ideology would be a second, unlabelled encoding of what the axes already say, and a caricature on top of that.

The quiz is an on-ramp to the sliders, not an instrument in its own right: it produces a starting position the reader is expected to argue with. It comes in four nested lengths — 6, 12, 20, 30 items — and every item, its axis and the pole agreement points at are published in the table below. Answers run −1 to +1 in steps of 0.5; each axis is the mean of its answered items after flipping each into axis direction. The mean rather than the sum is what makes the lengths comparable: the same opinions give the same coordinates whether answered over 6 items or 30, so a longer set buys resolution rather than distance.

Two constraints are worth stating plainly. Every length balances the three cube axes in both directions, because an axis measured from one side only measures agreeableness as much as opinion — which is why the shortest set is 6 items and not five. The 12-item set trades that balance on Democracy, Religion, Technology, Pace for breadth, and says so on the result; from 20 items on, every axis it asks about is balanced. An unanswered item contributes nothing rather than counting as a midpoint, and an axis with no answers behind it is left out of the distance entirely instead of being read as a zero the reader never claimed.

What is still missing is an uncertainty band. These items are editorial, unvalidated, and have no item-response calibration behind them, so the coordinates they produce carry a real error that the tool currently reports as a precise number. That is the honest next piece of work on this instrument.

Invented for this build: all 240 coordinates, the seven family groupings, the 0.12 tie threshold on the nearest label, and the wording of every axis definition. These are best-faith editorial priors, written to make the geometry teachable.

Needs real data before it can be trusted as measurement: the coordinates themselves (fittable to manifesto corpora, roll-call scaling or expert surveys), country and year medians for recentring, Overton bands, party positions, and any population density cloud.

Deviations from the build brief: the data model’s separate vec and radar fields were merged into a single 12-axis vector, because two overlapping copies of the same three numbers drift apart. Data is stored as typed TypeScript rather than raw JSON so a missing axis fails the type check instead of rendering as a hole in the radar. Routes are namespaced under /spectrum because this tool is a guest in a larger site and cannot own the root path.