The
The Science-based Assessment of Deep Personality
Documentation
Psychology runs on questionnaires, and nearly all of them are asked the same way: a page of radio buttons on a platform built for the person writing the study, not the person answering it.
The instruments are rarely what fails; the answering is. No amount of validation survives careless answers, and a dataset nobody can trust is not one worth sharing.
Qualtrics, Gorilla, jsPsych and the rest are good at building a study and indifferent to taking one: grid after grid, a progress bar at best, a thank-you at the end. Nothing in them is designed to be finished.
Attention falls with length. The same item is answered worse at the end of a long battery than at its start, and the people who stop halfway are not a random half.
Straight lines, the middle of every scale, a second an item: often a tenth of an online sample or more (Meade & Craig, 2012), in panels that pay by the minute and reward the quick. It is not only noise: it can make unrelated scales correlate (Huang et al., 2015).
An hour of honest answers buys a debriefing sheet. The participant learns nothing about themselves, so there is no reason to care how they answer, to come back or to tell anybody else.
One name for different things, different names for one thing. Scales multiply (most are used once or twice; Elson et al., 2023) and are seldom asked side by side, so how far grit is conscientiousness, or worry rumination, is rarely checked in one sample.
Each study asks a handful of scales of a few hundred people and keeps the file. Pooling them afterwards means reconciling versions, formats and samples that were never meant to meet.
A survey platform built to be engaging enough that people want to finish it. The questionnaires underneath are ordinary validated ones, asked as they were written; everything around them is there to keep somebody reading the question they are answering, to the last of them.
The aim is high-quality, re-usable, open-access datasets, clean enough to be pooled and wide enough to show how far their measures overlap, since the same people answer all of them.
The run is levels, dressed as a dive: a gauge in metres, landmarks passed on the way down, the seabed and the rock beneath it. What is left is always in sight and the next stretch always short.
Every level closes on a figure drawn from the answers just given (a body, a hill, a sea, a wheel) rather than a score. It is what is worth reaching, and why the next level gets answered.
After the first level, participants choose which part of themselves to test next, from two or three cards at a time. The order is drawn, then theirs; carrying on is their decision.
Each level finished mints a badge onto a shelf, and a whole-run profile fills in as the levels do. The run accumulates something to look at, not just a bar filling up.
Every reading can be agreed or disagreed with and every level rated. The verdict is data too, including on a star sign read beside a temperament: the Barnum effect caught in the act.
A level's results or the whole profile go out as a link or an image, which opens into the test for whoever receives it: recruitment by word of mouth, filed as such.
Answers are saved as they are given, and a run left partway is picked up where it stopped on the same device, so closing the tab costs neither the participant nor the data.
Attention checks, response times and a closing question on whether it was taken seriously are recorded for every level, and shown to nobody.
Built to be changed: the questions are data rather than code, and a study asks its own battery from its own link.