The Reasoner measures the structure of moral reasoning rather than the values a respondent endorses. For each dilemma it asks two things separately — what should happen, and why it matters — and has the respondent distribute a fixed pool of points across the answers. The same procedure scores humans and language models, so both land in one space. None of that machinery depends on which dimensions are being measured. The four axes of the pilot — Moral Agent, Authority, Moral Domain, Obligation Scope — are one choice of dimensions, not the instrument itself. Change the dimensions and the prompting, the running, and the scoring all carry over.
To test that, we pointed the instrument at a different set of dimensions: five adjudicative logics — the currency a person reaches for to settle a moral conflict. Welfare, the best overall outcome. Duty, what is right in principle. Honor, reputation and standing. The sacred, religious or divine obligation. And self-interest. This is a different shape of dimension from the pilot's — not four bipolar axes but a single competing set, where more of one logic means less of another. It dropped into the instrument without modification. And because the logics are fixed, they are the same five options under every dilemma: there are no per-scenario reasons to write, the dimensions carry the content. The only new code was a scorer that reads a five-way composition instead of a position on a line.
It does. Unframed, the models resolve almost everything by impartial welfare and duty, with honor, the sacred, and self-interest near the floor. Asked to answer as an ordinary person living in a particular country, the composition fans out — the sacred rises sharply under some framings, honor under others — and the models' own written reasoning shifts to match. None of this is a claim about the cultures named; it is the model's stereotype of them, and stereotype is the right word. The point is narrower, and it is about the tool: handed a dimension set it was never designed around, the instrument still moves, discriminates, and reads out something coherent — for models and, in principle, for people, on the same footing.
The prompt for this was small. A table in a passing post — four ways of adjudicating a moral conflict — made us wonder whether the Reasoner could measure something of that shape at all. A weekend's proof of concept later, it could, on the same runner that scores the pilot, with the logics written in our own words. We're grateful for the nudge.