What we will never claim

No diagnosis, no disorder, no condition, no assessment of mental health. Nothing about trauma, addiction, sexuality, criminality, or any characteristic a person can be discriminated against for.

This is not a limit we would lift if the model got better. Five friends describing somebody are not a clinical instrument, and a product that turned their observations into a health claim would be doing harm with a warm interface.

We will not tell you that you are in the top ten percent of anything, and we will not tell you your pattern is rare. Both need a comparison group large enough for the number to mean something, and ours is small.

Every product in this category publishes these numbers, which is worth thinking about, because most of them have the same problem we do. The day our base is large enough, this paragraph gets rewritten and the number that changed it gets published with it.

Not whether a relationship will last, not how somebody will do in a role, not what will happen next year. We have no outcome data of any kind, so a forecast would be a guess in a confident font.

Forward-looking material exists in one shape only: something to try this week, with an observable result, that you can report back as having done nothing.

You will never see a sentence one of them wrote. There is one exception and it is a single word supplied independently by three or more of them, which cannot be traced to any of them.

The reason to hold this line is not politeness. Somebody being honest about a person they love is doing something socially expensive. They do it once if it is safe, and never again if it is not.

Before your five answer, there is no panel, and a sentence beginning with the words people see you as would be a fabrication.

This binds our own marketing, including every type page on this site. Those pages describe a pattern and what it usually produces. None of them tells a reader what anybody thinks of them, because nobody has been asked.

The sixteen are corners of a continuous space and a person sits somewhere in it. The honest sentence is which corner is nearest, plus the runner up, plus a warning when the two are close.

A person near a boundary has been placed rather than diagnosed, and telling them otherwise would be the single most shareable thing we could do. The four questions explains what a placement actually is.

When you speak, audio becomes text and the recording is deleted within 24 hours. We do not model your voice, do not identify you by it, and do not infer emotion or personality from how you sound.

The interview engine cannot even tell which answers were spoken, because it only ever receives text. Not knowing is how the promise is kept, rather than a rule somebody has to remember.

A list of refusals is still a flattering thing to publish about yourself. Here is the harder half.

The test that would falsify this product is not whether readers agree with their Portrait. Forer already settled that: people agree with astrology filler at 4.3 out of 5. The test is whether somebody can pick their own Portrait out of a lineup of plausible wrong ones.

If subjects cannot do that at well above chance, the writing is Barnum prose however accurate it felt. The voice would be a failure regardless of how many people liked it. We think this is the right test, we intend to run it, and it can come out against us.

Two smaller ones. If a Portrait's findings do not survive the same five people being asked again months later, we are measuring mood rather than reputation. And if the gap between somebody's own account and their five is usually nothing, then the product has no subject matter and should say so.