How it works, and where it can be wrong

Transparency is the credibility. Here is exactly what the instrument does.

What it measures

Seven constructs across knowledge, morality, politics, and existence. Each is stored as one or more independent 0–100 sub-scores. Opposed pairs (e.g. Reason vs. Experience, free will vs. determinism) are shown as separate bars, so "high on both" — a Kantian, a compatibilist — reads differently from "undecided."

How it scores

Answers use a 1–5 agreement scale, centered so "neutral" contributes nothing (neutral = 50, not 60). Each construct includes reverse-worded items, so answering "agree" to everything cancels out instead of inflating every score — the flaw in the first version of this project. Where your answers within a construct disagree, we mark it tentative rather than pretending to precision.

What it is not

Data retention

By default your draft answers live in session storage until you finish. If you create a share link, a minimal server record (scores, consent time, optional integrity tap) is stored under an unguessable token for 30 days. You can clear browser data anytime via retake or closing the tab.

The AI layer

A preview interview UI exists; model wiring is post-pilot. When live, the interview will be bounded (re-weighting a numeric prior, not independent measurement), audited for documented LLM value-biases, and disclosed here in full.

Planned flow: surface one tension from your map → short follow-ups → steelman the opposing view.

Explore constructs

Per-construct explainers (SSR) describe what each axis means without taking the test.

← back