How this works
Axes are values and constraints, never correctness
Every instrument is built on a set of axes — how much structure you want, how you handle risk, how much consistency you can honestly sustain, how much you defer to expertise. None of them has a correct end. A tired parent on shifts is not wrong about screens; they have different constraints, and the instrument scores fit rather than merit. Nothing in the data model can even express "better".
A sourced matrix, with dispersion
Each approach has a position on every axis and a tolerated spread around it, with a citation. Gentle parenting is wide on structure — its own texts disagree with each other — and narrow on punishment, where they do not. Without that spread, an internally varied approach gets penalised for any answer away from its centre, which would be an artefact of the arithmetic rather than a finding about you.
Adaptive questions
The questionnaire asks whatever resolves the most and stops when the ranking is stable — usually nine to fourteen questions. Information gain alone would happily settle a result on four axes, so coverage is a hard constraint on top of it: a minimum number of axes established, quotas per group, and a couple of axes that must be asked outright. Every question describes a situation rather than a doctrine, and no option contains any camp's vocabulary.
Agreement, not a verdict
Each approach is scored independently on its own 0–100 scale, and the set is deliberately not normalised to sum to anything. Two approaches can both fit you at 88, and that is a true statement about them rather than a failure to decide. Normalising is exactly what manufactures "87% gentle, 84% Montessori", and it would also make your score depend on how many approaches happen to be in the matrix.
"Not sure" is a real answer
It is not a skip. An unsure answer leaves the ranking untouched and lowers the confidence of the axes it touched, and an axis you never took a position on is excluded from scoring entirely. We never impute a position you declined to take, and the result page names the axes you left open rather than quietly filling them in.
The report is generated, not written by a model
Every sentence in the paid report comes from the matrix and from copy written in advance. No language model touches it. That means it is traceable — a disputed claim can be walked back to a cell and the source that cell cites — reproducible, and free at the margin, which is what lets the free result be complete rather than a teaser.
What this cannot do
It cannot tell you what to do with your particular child, and it does not try. It measures internal consistency rather than real-world accuracy: the calibration runs simulated parents against the same matrix that scores them, which catches broken arithmetic and proves nothing about whether the rows are right. Only a panel of real parents who already know what they do can measure that, and until one has run, treat the result as a vocabulary for something you already knew rather than as a finding.
- 3 instruments
- 40 approaches
- 98 questions
- 52 sourced references