A single left-to-right line has to carry six independent questions at once: who owns things, how society should be ordered, how power is organised, at what level decisions are taken, who counts as a member, and by what means change is imposed. Compress six numbers into one and the information does not blur. It disappears.
This atlas keeps them apart. Every organisation in it holds a position on all six, so you can check whether two things called by the same name actually resemble each other — and where, exactly, they do not.
In May 1789 the Estates-General met at Versailles. Deputies who wanted to keep the royal veto sat to the president's right; those who wanted to abolish it sat to his left. The arrangement was furniture. It has been carrying the entire weight of political classification for two hundred and thirty-seven years.
Nobody designed the left–right line to measure anything. It recorded one binary vote in one chamber, and it happened to survive because it was memorable, because newspapers found it useful, and because a chamber can only have two sides. When Norberto Bobbio tried to give it a defensible core in 1994, he argued that the durable content is an attitude to equality: the left treats existing inequalities as remediable, the right treats them as natural or acceptable. That is a real distinction, and it is one distinction.
Everything else that gets loaded onto the line — attitudes to religion, to migration, to war, to the constitution, to the pace of change, to the legitimacy of force — has to travel as cargo. When the cargo conflicts with the label, the label wins, because it is shorter.
The academic response has been to add dimensions, repeatedly, and to find each time that the added dimension carries real independent information. Hans Eysenck proposed a radicalism axis crossed with a tough-mindedness axis in 1954. Milton Rokeach built a two-value model on freedom and equality in 1973. David Nolan's chart split economic from personal freedom in 1971, and the Political Compass popularised a version of it online from 2001.
The professional datasets went the same way. The Manifesto Project has coded party programmes since 1979 and its single left–right index, RILE, has been criticised for producing implausible jumps — parties apparently leapfrogging each other from one election to the next in systems that were not changing. The Chapel Hill Expert Survey, running since 1999, treats economic left–right and a cultural GAL–TAN dimension as two separate measurements, and its 2024 wave added items on executive constraints and judicial independence as a third concern entirely.
Michael Freeden's morphological account treats an ideology as a cluster of political concepts arranged in a structure: a core that all instances share, adjacent concepts that give the core a determinate meaning, and a periphery of policies that change with circumstance.
Ideological conflict is then a fight over decontestation — over which meaning of "freedom" or "equality" gets fixed. Two movements can share a word and share nothing else. A coordinate system cannot capture that structure, but it can stop you assuming that a shared word implies a shared position.
On the dominant account, associated with Cas Mudde, populism is a thin-centred ideology: it divides society into a pure people and a corrupt elite and insists politics express the general will, but it says nothing about ownership, borders or religion. It has to attach itself to a fuller ideology to say anything at all.
That is why left-populist and right-populist parties can share a structure and share no policy. Treating populism as a location on the spectrum guarantees a wrong answer.
Juraj Medzihorsky and Staffan Lindberg, working from the V-Party dataset covering nearly two thousand parties in 169 countries, argue that the trait which predicts democratic breakdown is anti-pluralism — and that it is empirically distinct from both populism and from ideological extremity.
The four signals come from Juan Linz: rejecting opponents' legitimacy, tolerating violence, denying the validity of elections, and curtailing civil liberties. None of them is a policy. This atlas gives them their own axis.
These are not accusations of bad faith. Most of them happen by accident, because the instrument is too coarse to say the thing the speaker means.
| Failure | What happens | Worked example |
|---|---|---|
| Collapse | Six independent measurements are reported as one number, so opposite positions land on the same point. | Pinochet's Chile and Kropotkin's anarchism are both "extreme". One privatised pensions under a military junta; the other abolished the state and money together. On the authority axis they are twenty points apart, the full width of the ruler. |
| Substitution | A name is treated as a measurement. The organisation's self-description is entered into evidence as its position. | "National Socialist" is a name. So is "Democratic People's Republic of Korea", "Liberal Party of Australia" — which is economically conservative — and "Republican", in countries where it means opposite things. |
| Drift | The midpoint moves while a position stays still, so a stationary organisation appears to have travelled a long way. | The British Conservative Party of 1975 supported nationalised rail, coal, steel and utilities. Measured against today's midpoint it reads as far left. It had not moved. The floor had. |
| Transfer | A word is carried across a border or a century into a frame where it does not hold. | "Liberal" means market economics in Europe and Australia, social democracy in the United States, and constitutional government in nineteenth-century Russia. Three different coordinates, one word. |
| Extremity conflation | Distance from the centre is treated as identical to willingness to abolish constraints on power. It is a different axis. | A party wanting a 90% top tax rate through legislation and a party wanting to disband the courts are both "extreme". Only one of them is a threat to the mechanism by which the other could be defeated. |
Each axis runs from −10 to +10 with zero as a genuine mixed or balanced position rather than an absence of opinion. The diamonds are real organisations from the atlas, coloured by era. Hover to read one; click to open it.
Why six and not two, or twenty? Two loses the distinctions that do the most damage when lost — above all, method of rule. Twenty produces axes so correlated that they are re-measuring each other. Six is the smallest set found, in building this, that keeps every pairing in the claim-check section legible without a footnote. It is a working choice, not a discovery, and the dataset is exported in a form that lets you rescore it or add a seventh.
Choose any two axes for the horizontal and vertical. Filter by era, kind of organisation and region. Marker shape shows what the organisation was — diamond for a regime that held state power, circle for an electoral party, triangle for a movement, cross for a doctrine. Size shows how much power it held. Dashed haloes mark lower-confidence codings. Every score in the atlas carries evidence; use the confidence filter to see only the entries coded at high confidence.
Selecting an organisation that moved draws its path through the space, with the year of each anchor. Nineteen have one. Try EC against AU first. It produces four populated corners rather than a diagonal smear, which is the whole argument in one picture: market economics appears at both ends of the authority axis, and so does collective economics.
Select up to four organisations. The profile overlays all of them; the distance figure measures the straight-line separation between the first two across all six axes at once. The maximum possible separation is 49.0 — two organisations at opposite extremes of everything.
The distance figure treats a ten-point gap on economics as exactly as important as a ten-point gap on method. Nothing justifies that except convenience. If you think the organisation of power should count for more than the scale of decision, say so here and watch which comparisons change and which do not. The equal-weight figure is shown alongside so the effect is always visible.
Ten equations that circulate constantly, taken seriously enough to be tested. Each gives the strongest case its defenders actually make, the strongest case against, and then what the coordinates show — which in several cases is that the claim is right about one axis and wrong about the map. Where the underlying dispute is conceptual rather than factual, that is said plainly instead of being resolved by assertion.
Paste a sentence that equates two political things. The checker finds the organisations named in it and reports the measured distance and which axes carry the disagreement. It matches names literally, so name both sides.
Twenty-four statements, four per axis. The result is not a label or a quadrant. It is six coordinates and the recorded organisations closest to them by measured distance — which is a more uncomfortable and more informative output than a colour. Nothing is transmitted anywhere; the scoring runs in your browser.
A word of caution that the instrument cannot enforce. Proximity here means similarity of stated position across six dimensions. It does not mean you would have joined, supported or excused anything. Half the value of the exercise is noticing that a movement you find repellent shares four axes with you, and that the axis it does not share is the one that mattered.
Every organisation plotted against the year it appeared. Choose an axis. The amber line is the median of a forty-year sliding window and the shaded band is its middle half — the range a contemporary observer would have experienced as the normal spread of opinion. The point is not that any single organisation travelled. Most did not. It is that the band moves underneath them, so the same coordinate reads as centrist in one decade and extreme in another.
Nineteen organisations recorded at four or five anchor periods each, plotted on any two axes and moved through time. Watch Fidesz cross the map from the Liberal International to the illiberal state, the Chinese Communist Party traverse the economic axis without moving on the authority axis, and Peronism visit both economic extremes under one banner.
Read the anchors, not the line. Each point is tied to a documented programme or governing act. The positions between anchors are interpolated by this interface and mean nothing on their own. No expert-survey series exists for most of these organisations before 1999, and none exists for the pre-war periods, so every trajectory is coded at low confidence. Hover any anchor in the atlas above to read what it is tied to.
—
Set the axis to EC — economic organisation and watch the band between 1945 and 1980. Nationalised utilities, price controls and full-employment mandates were the shared position of conservative and social democratic parties across Western Europe. Both major British parties of that period sit to the economic left of nearly every party in the 1989-onward cohort, including the ones now described as hard left.
Then set it to CU — social and cultural order. Here the drift runs the other way and much faster, which is one reason cultural questions now structure party competition in Europe more strongly than economic ones do. A position held without moving for forty years can travel the length of the visible spectrum without its holder changing a word.
Every organisation in this atlas has an ancestor, and position on the map says nothing about descent. Brothers of Italy sits fourteen points from Italian Fascism and carries its flame in the party symbol. Fidesz sits at the far end of the authority axis from the Liberal International it joined in 1990. The Kerala communists and the Vietnamese communists were founded from the same international and are now nineteen points apart.
Open any entry in the atlas and the panel now shows what it came from and what came after, each with the fact that establishes the link.
This atlas was built to be used on sentences you encounter in the wild. Here is the procedure, in the order that catches the most error for the least effort. It works without this site; the site only makes step three faster.
Ask what the claim is actually measuring. "Extreme", "radical", "hard right", "far left" are not measurements. Force the sentence to say whether it is about ownership, social order, the organisation of power, the level of decision, who belongs, or method.
Most claims collapse here, because most of them turn out to be about one axis while being stated as if they were about all six.
A party's name is evidence of what it wanted to be called, not of what it did. National Socialist, Democratic People's Republic, Liberal Party of Australia, Institutional Revolutionary Party. Every one of those names points somewhere other than the position.
The test is what the organisation legislated, expropriated, banned or built — not what it printed on its banner.
Load both things into the comparison. The number matters less than which axes carry it. Two organisations can sit the same distance apart and be apart for entirely different reasons, and the reason is the whole content of the claim.
If the divergence is concentrated on one axis, the claim is a statement about that axis. Calling it anything larger is the error this atlas exists to catch.
Ask when the comparison is being drawn from. The British Conservative Party of 1975 supported nationalised rail, coal and steel. The National Rally campaigned on Reaganite tax cuts in 1984. Fidesz was a member of the Liberal International.
An organisation that has not moved can appear to have travelled a long way, because the midpoint moved underneath it. Run the trajectory before you assert continuity or betrayal.
Run the procedure on the claim your own side makes most confidently about its opponents, before you run it on theirs. If the answer is one axis out of six, you have found the same error you were looking for elsewhere.
The failure modes in section I are not partisan. Substitution is used to prove that Nazis were socialists and that market liberals are fascists. Collapse is used to equate social democrats with Bolsheviks and constitutional conservatives with the Klan. The instrument is symmetrical because the mistake is.
Version 1 of this atlas asserted 1,068 numbers and sourced none of them. That is the same failure the whole project exists to catch, so it is fixed here in four parts: a published codebook, an audit of the instrument computed live from its own data, a validator that checks the atlas against real professional datasets, and a mechanism for disputing any single score.
What each score is supposed to mean, fixed with anchor cases so that coding is calibrated rather than intuited. This is the technique the Chapel Hill surveys use to keep hundreds of separate coders on the same scale.
Everything below is computed from the dataset when the page loads. If the data changes, these numbers change, including in directions that are unflattering. Two of the three panels currently report findings that weaken the atlas rather than support it.
Before any argument about what the numbers mean, a reader is entitled to know that the file is not broken. These are the checks that ran in my terminal after every edit while this was built. They now run here instead, on your page load, so that the answer does not depend on my say-so.
Since version 2.0 the audit has reported that two of the six axes correlate at around 0.87, and the atlas has answered by asserting that the cases where they diverge are the ones that matter. That was an assertion. Below is the test: build the five-axis model the objection proposes, recompute every pairwise distance in the dataset, and report what changes. It runs on page load and the result is not flattering.
The previous test asked whether two axes could be collapsed into one. This asks the same question of all six: drop each in turn, recompute every pairwise distance, and measure how much of the map survives without it. The aggregate answer is not the one the atlas would have chosen.
The strongest objection to this atlas is that its numbers came from nowhere. So test them. Download a CHES trend file from chesdata.eu or a V-Party export from v-dem.net, export it as CSV, and drop it here. The site will align the party names it recognises, compute its own agreement with the professional measurement, and list the cases where it disagrees most.
Expected columns: lrecon and galtan for CHES, v2xpa_antiplural and v2pariglef for V-Party. CHES scales run 0–10 and the anti-pluralism index runs 0–1; both are rescaled to the atlas range before comparison, and the conversion used is printed alongside each result. Matching is on party name after normalisation, with an alias table for common abbreviations, so coverage of any given file will be partial and the matched count is always shown.
Open any entry in the atlas, expand the evidence panel, and challenge the cell you think is wrong. The form produces a structured record with the current score, your proposed score and your evidence, ready to submit. A challenge without an evidence field is triaged as a preference, which is a different thing and is treated as one.
A documented act, a clause of a programme, a statute, a court finding, or a published measurement from a comparative dataset. Anything a third party can go and check without taking your word or ours.
That the score feels too harsh or too generous. That the organisation is worse or better than the number implies. That a different organisation scored differently. Those are arguments about the world, not about the coding, and they belong in the claim check rather than here.
The v2.0 audit reported the two cultural axes correlating at 0.88. That prompted a review of the membership axis — and reviewing a measurement because a statistic embarrassed you is a real bias risk, so the procedure was fixed before any score moved: sharpen the definition first, apply it mechanically to every entry, publish the resulting correlation whatever it turned out to be.
The rule that changed: the membership axis was being coded from immigration policy. It should be coded from the rule of membership. Wanting fewer arrivals is a claim about numbers; defining the nation by descent is a claim about who can ever belong. A party that would admit nobody new but treats every naturalised citizen as fully equal is restrictive on the border and civic on belonging, and most of the sample was being scored as though those were the same thing.
The correlation moved from 0.883 to 0.868. Thirty corrections, and the statistic barely noticed — which is itself the finding. Corrections are applied at load rather than edited into the data, so the original score is preserved and every change below can be checked or reversed.
An atlas that revises itself in public is the only kind whose epistemics match its argument.
An instrument built to expose overconfident classification has an obligation not to be overconfident. Here is exactly what was done and where it breaks.
Each organisation was scored on all six axes against its programme where one exists, its record in office where it held office, and the settled findings of the comparative-politics literature where the two disagree. Where a movement changed substantially, the period stated in the record is the period coded — Sandinismo is coded post-2007, Peronism and Kirchnerism are separate entries, Deng-era and Xi-era China are separate entries.
Confidence is recorded on every entry. Low confidence means either that the organisation is recent and moving, that reliable independent reporting is thin, or that qualified specialists disagree about the classification itself rather than about the facts. Entries marked contested are ones where the disagreement is the point, and both readings are given.
| EC | Economic left–right as operationalised by the Manifesto Project and the Chapel Hill Expert Survey. |
| CU | The GAL–TAN dimension from the Chapel Hill surveys; adjacent to the Inglehart–Welzel traditional/secular and survival/self-expression axes. |
| AU | Linz's litmus test for semi-loyal parties, operationalised as the V-Party Anti-Pluralism Index, plus the executive-constraint and judicial-independence items added to the 2024 Chapel Hill wave. |
| SC | Centre–periphery and territorial dimensions in the comparative federalism literature, and the subsidiarity principle in Catholic social thought. |
| ID | The civic/ethnic distinction in nationalism studies, and the V-Party items on immigration, cultural superiority and minority rights. |
| CH | The reform/revolution distinction internal to socialist historiography, combined with the political-violence component of the Linz criteria. |
1. It is not objective, and no coding scheme is. Expert surveys built by teams of hundreds of political scientists produce measurable disagreement between coders on exactly these questions. A number rendered in monospace still came from a judgement. What the coordinates buy you is not certainty — it is a shared, inspectable statement of what is being claimed, which an insult does not provide.
2. It is a synthesis, not a peer-reviewed dataset. The scores here were assembled from the published literature and from the public record of each organisation. They are not the output of the Chapel Hill, Manifesto or V-Party projects, and should not be cited as such. Those projects publish their own data; where you need authoritative figures, use theirs — and use the validator above to see how far this instrument is from them.
2b. Full coverage is not the same as being right. Every cell now carries a rationale and something citable, which removes the weakest objection to this atlas and leaves the strongest one standing. An evidence item can be present, genuinely checkable, and still be the wrong item — selected because it supported the score rather than because it was the most important fact about the case. No audit detects that. The codebook, the external validator and the challenge mechanism are the only defences, and all three depend on someone bothering to use them.
3. It is a sample, not a census. Three hundred and eighteen organisations are recorded here, weighted toward those that recur in argument. There are thousands of registered parties in the world. Absence from this map means only that it is not here.
4. It says nothing about whether a position is right. A coordinate is a description. Two organisations can be adjacent on the map and one of them can be catastrophic in practice. Distance measures similarity, not merit, and no arrangement of axes will tell you which programme to support.
5. It cannot settle legal or moral questions. Whether a particular state's conduct meets a legal definition — of apartheid, of genocide, of a coup — is a question for courts and evidence, not for classification. A map of ideological positions is the wrong instrument and will give a confident wrong answer if you force it.
6. It ages. Contemporary parties move. Several entries here are coded to a moment that is already passing. Check the year on the entry before you rely on it, and export the data and rescore it if you have better evidence.
The failure modes in section I are not partisan. Substitution is used to prove that Nazis were socialists and to prove that market liberals are fascists. Collapse is used to equate social democrats with Bolsheviks and to equate constitutional conservatives with the Klan. The instrument is symmetrical because the error is.
The most useful test is the uncomfortable one: take the claim your own side makes most confidently about its opponents, load both coordinates, and look at which axes actually diverge. If the answer is one axis out of six, the claim is a measurement of that axis, and calling it anything larger is the thing this atlas exists to catch.
The JSON export carries everything this site knows: every organisation, all six coordinates, the full evidence layer with its rationales and artefacts, the trajectories with their anchors, the revision log, the codebook, the schema and the caveats. The CSV is the flat coordinate table for spreadsheet work. Rescore it, extend it, argue with it, or plug it into something else.
Primary sources worth going to directly: the Chapel Hill Expert Survey (chesdata.eu), the Manifesto Project (manifesto-project.wzb.eu), V-Party and V-Dem (v-dem.net), and the Global Party Survey. Michael Freeden's Ideologies and Political Theory for the morphological account; Cas Mudde on populism as a thin-centred ideology; Juan Linz on semi-loyal parties; Norberto Bobbio's Left and Right for the strongest defence of keeping the single line.