← All articles

Inside the Free Scan: 92 Items, One Algorithm, and the Honesty of Stopping

We give away a personality scan at /test. It takes about two minutes, asks at most 30 questions, and usually fewer. This article is the complete engineering behind it — the item bank, the algorithm, the stopping rule, and the simulation numbers, including the ones that show where the instrument's limits are.

I'm publishing the internals for one reason: a measurement you can't inspect is an opinion with a progress bar. Ours is a machine, and machines can be opened.

What the scan measures — and what it deliberately doesn't

The Ten Aspects system describes ten functional energies — Crown, Amok, Seal, Crest, Epicure, Maverick, Wealth, Windfall, Twin, Plunder. Every person runs on some mix of them, and almost everyone has one or two doing most of the work.

The scan answers exactly one question: which energy are you running on right now? Not which one you were born holding — those are different facts. A person can be born a Seal and run on Amok for a decade because life demanded it. The questionnaire reads the runtime; the birth chart reads the inventory. We keep that boundary explicit because most tests in this genre quietly promise both and deliver neither.

The bank: 92 items, ten ledgers

Under the scan sits a bank of 92 statements. Each one is a small, deliberate instrument with three parts:

• A target aspect. Every statement reads one energy and only one. About nine items per aspect.
• A polarity. Some items are keyed forward ("agreeing" adds evidence), some are keyed in reverse — agreeing subtracts. Reverse-keyed items exist because people who want a certain result will agree with anything that sounds like it. Reversed items catch the performance.
• A weight. Not all statements are equally diagnostic. An item that only a genuine Amok would endorse carries more weight than one that half the population nods along to.

The statements themselves are behavioral, not aspirational. We don't ask "are you competitive?" — everyone has a flattering answer to that. We ask about what you actually did: the situation you exited, the thing you counted, the moment your hands got steady. Behavior is harder to fake than self-image, and when it is faked, the reverse-keyed items usually collect the evidence.

You answer on a five-point scale, from strongly disagree to strongly agree. Every answer updates a running score on the target aspect — answer × polarity × weight. Ten ledgers, updated live.

Why adaptive: certainty is hunted, not accumulated

A fixed 92-question test wastes most of its questions. If your first four Epicure answers all came back flat, asking the remaining five Epicure items teaches us almost nothing — the ledger has already spoken. Fixed-length tests keep asking anyway, because their length was decided before you sat down.

Ours decides its length while you answer. The scan runs a loop:

Phase one — the spine. The first ten questions are fixed: one anchor item per aspect, the highest-weight forward-keyed statement in each ledger. Everyone gets these same ten, in the same order. This is deliberate — before the machine can adapt, every aspect deserves one honest reading. No energy gets eliminated on zero evidence.

Phase two — feed the contenders. From question eleven, the machine looks at the current ranking and asks: who is still contesting the top? Aspects within striking distance of the leader are contenders; everyone else has structurally lost and stops receiving questions. Among the contenders, the one with the fewest readings gets the next item — that's where one more answer buys the most information. If your top is a fight between Seal and Crest, you'll get Seal and Crest questions. Someone else's fight between Wealth and Windfall gets Wealth and Windfall questions. Two people can take "the same test" and, from question eleven, see almost none of the same items.

Phase three — the gate. After every answer, three conditions are checked: at least 18 questions asked; the leader's normalized margin over #2 at or above a fixed threshold; and both of the top two aspects read at least three times each. All three pass — the scan stops and names your engine. Any fails — it keeps asking, up to a hard cap of 30.

The cap matters for a specific reason: boundary cases get more questions, never a coin flip. If your top two are close, the machine's response is to keep gathering evidence, not to shrug and pick one.

Why the same answers always give the same result

If you retake the scan and answer the same way, you will see the same questions and get the same result — every time. People sometimes read this as "not really adaptive." It's the opposite: the scan adapts to your answers, not to a random number generator.

This was a design decision with a name: the anti-whiplash rule. A test that shuffles its items can hand the same honest person three different identities in three sittings — at which point the person rightly concludes the test measures nothing. Determinism is what makes the instrument falsifiable: same input, same output, inspect it all you like. The first ten questions are identical for everyone by design; what changes between people — and between your honest self and your performed self — is everything after.

The simulation: what 30 questions actually buy

Before shipping the cap at 30, we simulated thousands of test-takers. Each simulated person got a true affinity for every aspect, answered every question honestly-but-noisily — the way real humans do — and took the test two ways: the full 92-item grind, and the adaptive scan with its cap. Then we compared verdicts.

Three findings, one of which changed the product.

First: when a person has one dominant engine, the scan nails it. Clear-leader profiles agree with the full 92-item bank about 94% of the time at the 30-question cap. For the thing the scan is for — naming your primary energy — 30 questions carry essentially all the signal that 92 do.

Second: raising the cap buys almost nothing. We ran the same simulation with caps of 36 and 40. Overall agreement moved by about two percentage points — for ten more questions of everyone's patience. The bank has ~9 items per aspect; by question 30 the contenders' high-weight items are mostly spent, and further questions are low-information by construction. The cap stays at 30.

Third — the finding that changed the product: near-ties are real, and no test length fixes them. For simulated people whose top two affinities were genuinely close, the adaptive scan matched the full bank's #1 only ~42% of the time. But here is the detail that matters: the full 92-item bank couldn't rank those people stably either. Run it twice with fresh noise and it flips its own verdict. That's not measurement error. That's a fact about the person: they run two engines in parallel, and the honest reading of a tie is a tie.

So that's what the scan now reports. If your top two land within a few points of each other, the result page doesn't force a winner — it names you a dual engine, shows both totems, and gives you both manuals. A test that pretends to a precision it doesn't have is a horoscope with extra steps. The near-tie is the finding.

Where the questionnaire's floor is

Every questionnaire — ours included — has a hard ceiling, and it's worth stating plainly because it defines what the scan can and cannot claim.

A questionnaire reads behavior, and behavior is the output of a life, not the inventory. The scan cannot distinguish an energy you were born holding from one that was trained into you by a decade of circumstances, or one that's on loan from the ten-year cycle you happen to be walking through. All three produce the same answers to the same 30 questions, because all three produce the same behavior — the differences are in the origin and the expiry date, and those aren't visible from inside a sitting.

That boundary isn't a marketing hedge; it's where one instrument ends and another begins. The scan reads the runtime. Reading the inventory — what's native, what's installed, what expires and when — takes the birth structure, which is a computation over a different input entirely: the time coordinate, not the self-report. That machine is described elsewhere in these articles.

One instrument, one claim. The scan claims the runtime, shows its work, and stops asking the moment it knows. That's the whole design.

— Daozi Wu, Founder

Wu, D. (July 19, 2026). Inside the Free Scan: 92 Items, One Algorithm, and the Honesty of Stopping. DaoTiming. https://daotiming.com/articles/inside-the-ten-aspects-scan