Publishing the arithmetic
Every formula this site uses is written out on a page anybody can read, including the ones that make it look worse.
crafthonestymethod
Notebook
Notebook
A development blog for this project — why the question is worth asking, what the prior work already settled, what these measurements can and cannot support, and what is still unknown.
These were written after the fact, from the commit history and working notes. The dates are when the work happened, not when the entry was written. Nothing here is a changelog: where something went wrong, the entry is about the idea it exposed.
New here? Start at the bottom, with The number that described nobody — the profile that set this off. The list below runs newest first.
What I still do not know
The list of things that could make this wrong, kept where I have to look at it.
open-questionshonesty
The dataset is the point
The individual results are the thing that gets people to answer. The dataset is the thing that might matter.
dataaspirationopen-science
Saying "I cannot tell"
Nine in ten of the traits this instrument called steady were gradients it simply could not resolve.
measurementhonesty
Borrowing 619,150 people
A percentile needs a reference sample. I have no lab, so I used somebody else’s — and it doubles as the largest test in the project.
datamethod
Every number gets an error bar
A four-item facet carries about fifteen percentile points of noise, and nearly every test prints a bare number anyway.
measurementhonesty
Your score is partly luck
The number this site leads with is not the percentile of your score. It is deliberately pulled toward the middle.
measurementmethod
The rule that keeps this from being astrology
Four points always look like they have a shape. That is true of noise too.
measurementhonesty
Drawing at the resolution the sample supports
Binning a distribution invents shape. So does refusing to bin one — it just looks more honest while doing it.
craftcharts
It is not one number
The thesis is that a spread exists, not that it runs in any particular direction.
thesis
Why four rings and not two
The obvious design is to ask twice and subtract. The obvious design does not work.
methodpsychometrics
What a control is for
Three of the ten traits are predicted not to move at all. Without them the hypothesis cannot fail.
methodfalsifiability
The facet I refused to predict
One of the ten carries no prediction at all, because I genuinely do not know which way it should run.
methodfalsifiability
The paper that could kill this
In 2023 somebody ran almost exactly this manipulation and found essentially nothing.
prior-artfalsifiability
Somebody got here first
The closeness gradient was reported in 2014. Being clear about that is the only way the rest is worth reading.
prior-art
A type system that admits what it is
I built a five-letter code on a site that argues against exactly that, and then made it tell you how firm it is.
typologyhonesty
The reflex I could feel myself having
Modesty at the 97th percentile is a real score and a measurement artifact contaminating everything around it.
originresponse-style
I went back and read the items
Altruism, 33rd percentile — for someone who has spent a large part of his adult life caring for people.
originitemsthesis
The number that described nobody
My Neuroticism came back at the 7th percentile, which is true of five of the six things underneath it.
originfacets
The instrument itself is at /methods, every formula is written out at /math, and the conditions under which this project would consider itself wrong are at /why.