How your result is matched to the library

Each profile is stored as twelve values

The library holds 31 ideologies, 28 modelled civic traditions and 53 historical figures. Every one of them is stored with a single recorded value for each of the twelve axes — the same twelve values your result produces — and those stored values are what a result is compared against.

The stored vectors are editorial readings of the recorded positions, written by hand and open to disagreement. They are not measurements of any party, movement or voter, and the library prints no percentage for the distance between two of its own entries.

Direction only, on one shared curve

The comparison is direction-only: it asks whether two twelve-value patterns point the same way, not whether one was answered more firmly than the other. Someone who agrees mildly with the whole of a profile can read as closely as someone who agrees strongly, because intensity is not what the ranking is measuring.

Both sides of the comparison are normalized before they are compared, and the result is placed on a 0–100 curve anchored so that one chosen level of agreement reads as 50. The same curve is used whether your result is being ranked against the library or compared with a friend's, so the two numbers mean the same thing.

When the test declines to give a number

If the answers carry too little overall direction, no match is reported at all — the page shows the empty state instead. A percentage in that situation would be decided by rounding noise, and printing it would imply an agreement that is not there.

Below the halfway mark the copy says the library has no close neighbour for the pattern rather than describing the result as a close fit. The percentage itself is printed unchanged; only the sentence around it changes.

Ranking follows the percentage. Where two candidates print the same number, the tie is broken by whether one of them leans against the axis you answered most strongly.