Analysis, not guesswork
There are two ways to build something like Valoo. The first: feed thousands of examples to a model, let it find the patterns itself, and accept whatever comes out. The second: write the rules one at a time, and make sure every rule can be opened and inspected. Valoo took the second, and that single choice determines almost everything else on this page.
You feel the consequence most when something looks odd. Point at a line in your report and ask “where did this come from?” and there's always an answer, by a short route: from the sentence back to the score, from the score back to the item you answered. No part of it can only be explained with “that's just what it produced”.
There's no machine learning anywhere in the decision path — not a little, none. AI is still used in Valoo, but it works at the other end and its share is far smaller than people usually assume. The last section of this page sets out exactly how far it goes.
97 answers, read from several angles
The Persona Compass uses 35 items — 7 for each of the five sides of character. Seven isn't a comfortable number, it's a necessary one: a trait measured by only two or three questions wobbles the moment you misread a single sentence, and a result that wobbles on the last question isn't fit for choosing an income path.
Beyond those 35 items, Valoo reads where you stand today from other angles: 12 personas · 5 spectra · 8 capitals · 8 income paths. 97 things you answer in total. One sitting or several — the draft is saved on your own device, so being interrupted never means starting over.
Why some statements are written in reverse
15 of the 35 persona items are deliberately written in reverse. If you're genuinely considering each statement, your answers stay consistent even as the direction flips. If you're speed-ticking one column to get it over with, your answers start contradicting each other — and the engine reads that contradiction.
It's the cheapest careless-answer detector there is, and the hardest to game — because gaming it requires reading every statement carefully. And someone reading every statement carefully is answering properly anyway.
The sentences are assembled, not pulled off a shelf
The easiest way to build a personality report is to write twenty paragraphs and pick whichever comes closest. It looks tidy and it feels personal — while what you're reading is one of twenty paragraphs everyone else gets too.
Valoo assembles your sentences from your own numbers, and the assembly rules carry one hard requirement: every sentence has to name something that isn't universally true. The test is simple and it's genuinely applied — swap the persona name and the sentence must immediately become wrong. If it still reads as true after the swap, it doesn't ship.
The same principle applies when Valoo reads your weak spot. The lowest capitals are read as a pair, not singly, so a thin margin never hardens into an accusation. And if all your capitals come out high, Valoo doesn't go hunting for a weakness — not everyone has one worth naming, and inventing one to fill a slot is the exact opposite of what this tool is for.
Your answers never leave the engine
Valoo uses AI for exactly one thing: smoothing sentences so they read well. It doesn't calculate, it doesn't select, and it has no part in any decision.
And what it receives isn't your answers. Only five pieces of already-computed output cross over: persona, path, mode, action code, and tone. Your raw answers, your income figures, your written reflections, and your name never make that crossing.
What makes that promisable isn't good intentions — it's the shape of the door. The door to the AI only fits those five pieces, so a sixth can't slip through even if someone writes the code wrong. Two layers of checks guard it, plus an automated sweep before every release that blocks the release if anything gets past.