The trait profile uses the IPIP-NEO-120 (International Personality Item Pool; public domain), 120 questions covering five domains and thirty facets, four questions each. The state check-ins use the PHQ-8 and GAD-7 (Pfizer, released without restriction), displayed under the names Low mood and Worry, plus three questions of our own (Stress load, Life satisfaction, Connection) which are original wording and not normed. Question text and answer choices are administered verbatim; we never edit an instrument.

Trait percentiles are computed against tables we derived from Johnson's published IPIP-NEO-120 dataset (OSF; 339,876 complete adult cases), banded by sex and age group (18–24, 25–34, 35–49, 50–90). If you didn't state a sex or age, you are compared with all adults. A percentile of 81 means: higher than 81% of the adults in your comparison group. This sample is people who chose to take a long personality test on the internet, not a random sample of all adults. The same is true of every free Big Five test. Norm data v1.0.0.

During the first assessment you get an early read at 30, 60, and 90 answers, shown as ranges rather than single numbers. The point estimate is your answers so far, scaled to the full questionnaire and run through the same norm tables as everything else. The range around it is empirical: we took all 339,876 complete cases in the source dataset, cut each one off at exactly those checkpoints in our question order, and recorded how far the early read landed from the final one. The range shown is the middle 80% of those real gaps, for people whose early read looked like yours. On the source data the final score lands inside the range about 82% of the time at 30 answers. Ranges are display only, and never feed the final report. Calibration v1.0.0.

Every “In words” passage is written by fixed rules, not by a model, and it only restates what the numbers already say. Position words follow set cut points on the percentile scale: near the top from the 85th, toward the top from the 70th, around the middle from the 31st, toward the bottom from the 16th, near the bottom below that. A split inside a broad trait is called out when its highest and lowest facets are 30 or more points apart. Baseline passages repeat the baseline, typical range, variability word and any moved readings exactly as computed, and name the steadiest and most variable of your numbers by wobble relative to each scale. The passages keep the numbers out of the prose on purpose; the bars and rows beside them carry those. The overview uses no trait names at all: it leads with whatever sits furthest from the middle, in the same words, says the rest sits close to most people, and then, only when two parts of one trait sit clearly apart, puts the contrast in plain words. Each broad trait gets three sentences: what the scale is about, where you come out in the scale's own words for that direction (the same words its definition uses for higher and lower scores), and which of its six parts show up most and least in your answers. Explanation stops there. What any of it means for your life, your work or your relationships is never stated, because the questionnaire cannot know it. Nothing is inferred about you and nothing is advised. The same numbers always produce the same sentences. Words v1.4.1.

Your baseline for each number is the middle (median) of your own completed check-ins. Your typical range is that middle plus or minus your usual wobble (the median absolute deviation), and never narrower than ±1 point on the 0–10 sliders and ±1.5 on the two bigger scales, so an ordinary answer on a steady week can't look like an event. A check-in counts as moved when it lands more than 3 points outside the range on the bigger scales, or more than 2 on the sliders. These rules were fixed before any user data existed and are versioned with the engine.

Every check-in is checked four ways: time taken (a full assessment under 8 minutes, or a short check-in under 3, is flagged), identical answers in a row (flagged at 8), a catch question that is almost never true, and pairs of questions that ask the same thing in opposite directions (flagged when they disagree by 2 or more). A flagged check-in is kept and counted, and the flag is printed in your report. Nothing is quietly dropped.

A few of the questionnaire's official facet names are words we keep out of this product on purpose. Where we renamed one, the mapping is here. The questions and scoring behind each name are unchanged.

Scoring is a fixed, versioned set of rules, with no AI anywhere in the path. Engine v0.1.1. Every answer you give is stored in an append-only log, and your reports can be recomputed from it exactly. Export or delete everything from your account page.