CoreChoice

I went looking for a proper Big Five test — the fifty-item kind with real psychometric work behind it — and kept finding the same shape. Ten minutes of honest questions about yourself, then a payment screen standing between you and your own answers. Finding one that simply gave you the result was harder than it has any right to be. So the test here is free and complete, and finishing it hands you five analyses on top.

The rest of it is what I wanted the profile for in the first place. When a decision has been going round your head for a week, another list of pros and cons is not the thing that is missing. What helps is someone thinking it through with you who knows how you tend to decide — one recommendation, the reasoning laid out, and a straight this is too close to call when it is. The answer screen closes on a line doing more work than it looks: you can sit with this, it will be here in the morning.

There is one place I ask for money. Each analysis is a call to a model and that costs me real money to serve; the test, your profile and the result cost me nothing, so they are free. That rule is most of why I build any of this — the things I went looking for and could not find, made so that somebody else’s week is easier than mine was.

That the result is free is not a promise in the marketing copy, it is a shape in the code. The screen showing your profile has no dependency on the coin ledger at all, so a paywall could not be added there without rewriting it first. The result renders on a phone with no signal, because nothing about showing you your own answers needs me.

Nothing you type is stored on the server. The dilemma and your trait scores travel in the request body, are held in memory long enough to build the prompt, and are gone — what persists is a coin balance against a random identifier, a salted IP hash for abuse limits, and a usage row for cost tracking. There is no account, no email and no sign-in anywhere in the app.

The third rule is the one nobody notices when it is working: a coin is never lost to a failure of mine. Every path that spends one puts it back if anything downstream falls over — the model timing out, a response that cannot be parsed, even the phone hanging up mid-request. Two bugs of exactly that shape surfaced during the build, which is why it is now held in place by tests rather than good intentions.

The measurement is the real thing: the fifty-item IPIP Big-Five Factor Markers, which are public domain, with the reverse-keyed items kept honest by tests that fail if the keying drifts. Where IPIP has published an official translation of those fifty items, that is what you answer. Where it has not, you answer mine — marked as mine, on the screen, with the original English a tap away on every question. I would rather tell you which one you are taking than quietly imply they are the same thing.

The scores never reach you as percentages either. A number implies something is missing out of a hundred, and to someone already prone to over-deciding, “75%” reads as a one-in-four chance of ruining their life — so four plain labels replace it, and a trait is drawn as a point on a line rather than a bar that looks part-empty.