Transparent by design
How the Atlas draws a map.
No language model decides your coordinates. Every option contains authored weights, and the same answer set always produces the same result.
Three coordinates, kept independent
Each selected option can contribute from −3 to +3 on outcomes versus principles, universal versus contextual morality, and impartial versus relational morality. Scores are normalised against the questions you actually saw, then placed on a −100 to +100 range.
Confidence changes weight, not ideology
Your one-to-five confidence rating applies a multiplier from 0.6 to 1.4. It can strengthen an answer, but never changes what the answer represents. Unweighted and confidence-weighted results are both retained.
Later questions target uncertainty
Twelve balanced core dilemmas establish an initial map. Later questions are chosen deterministically from the next chapter to add information about the least-supported axis while avoiding repeated families.
Tensions are not treated as hypocrisy
Matched dilemmas vary one meaningful detail, such as whether the person at risk is a stranger or someone loved. A changed answer is reported as a contextual turn or tension. Influence is measured by recalculating the map without each answer.
Reference points are interpretive
Philosophers and traditions are located using published scholarship and deliberately wide uncertainty. Many views fit one axis poorly; a coordinate near zero can mean “not well represented,” not simply “balanced.”
For the full formulas, data assumptions and limitations, see the project’s documented methodology.
Take the atlas