How the questions are built
A belief map depends on the questions behind it. Two common sources of bias matter here:
- Acquiescence bias: people may agree with whatever statement is put in front of them.
- Framing bias: people may respond to the wording of a question rather than the underlying belief.
The Belief Genome Project aims to account for both. Every belief dimension has 12 questions, and an automated check verifies that structure.
Equality pairs: ask both ways with equal weight
Each belief is asked twice: once leaning toward it ("canonical") and once leaning against it ("inverted"). We call these equality pairs. Both halves carry identical weighting and use the same frame. Only the direction differs, so neither side is the "easy" answer.
Toward: "An external physical reality exists whether or not anyone is there to perceive it."
Against: "Reality is a construction of the mind — without observers, the concept loses meaning."
Because every belief is asked in both directions, anyone who simply agrees with everything will contradict themselves. The system can flag that pattern. This helps separate consistent responses from a tendency to agree.
Six angles per belief
Asking a belief one way only reveals how someone reacts to that phrasing. So each belief is framed six different ways, and each frame is asked in both directions:
| Frame | What it tests |
|---|---|
| Standard | the plain proposition |
| Personal | lived, first-person experience |
| Societal | the social / collective angle |
| Edge case | a demanding, specific test |
| Concrete | a vivid, real-world scene |
| Abstract | the underlying philosophical core |
This is triangulation, not repetition. A belief that holds in the abstract but collapses in a concrete moment (or vice-versa) exposes a real signal: how consistently the belief is held. That consistency is encoded directly into the Belief Genome.
Enforced, not promised
The balance is checked by an automated validator that runs across the entire question bank whenever it changes and fails if any rule is broken:
- Each dimension has exactly 6 pairs, with an exact canonical/inverted balance.
- Every pair is one canonical + one inverted question targeting the same dimension.
- Both halves of a pair carry equal weighting and the same frame type.
- All six frame types appear in every dimension, evenly distributed.
Current bank
124 dimensions · 1,488 questions · 744 pairs · 248 questions per frame type · 0 balance errors.
What we claim and what we don't
This methodology guarantees structural rigor: balanced pairs, even framing, and complete coverage across all 124 dimensions. It does not, on its own, claim empirical validation, or proof that these dimensions predict behavior outside the test. That is ongoing work. We claim the discipline that makes such validation possible, and we report our figures transparently and reproducibly.