How they move is the signal
Every reach is recorded, from how straight the path is and the small corrections on the way to where the top speed falls and the natural tremor of a human hand. People have a signature, and scripts do not.
Miraai Reasoning
A reasoning test made to make it harder for candidates to cheat.
Since candidates started keeping an AI open in the next tab, a reasoning test has been hard to trust. We built one that makes cheating harder by design, because the question stays on screen only while the candidate holds it there.
Three example items, live, with the interaction trace beside them. Nothing is scored and nothing is recorded. On a phone, turn it sideways.
What you can rely on
Harder to cheat by design, with proctoring when you want it.
Any reasoning test can now be answered by a phone in the candidate’s other hand, and most vendors respond by hunting for questions AI cannot solve, a race that is lost within months every time. We made getting help slower than doing the work instead, because the question only exists while the candidate is holding it and the clock does not care why they let go.
The test also adjusts to each candidate, so a strong applicant is not bored by easy questions and a weaker one is not buried by hard ones, and every score sits on the same scale whichever questions a person saw. New questions are calibrated quickly and the pool is refreshed regularly, so nothing stays in circulation long enough to leak.
Run it on its own, or add facial validation and optional AI video proctoring, a flag for a person to review, on Podium365 when you also want to rule out a stand-in or a helper.
Held and released
Letting go costs the question, not the time.
The clock starts when the candidate first holds the orb on an item and then runs without stopping, never speeding up and never pausing, until it runs out or the next item begins.
The interaction trace
Every reach is read, and a person makes the call.
It feels different every time
The orb behaves slightly differently each session. A person adjusts within a few reaches, while a script replaying a learned movement either does not adjust or adjusts wrongly.
A flag, never a verdict
The check returns a score and the reasons behind it for a person to review, and no decision about a candidate rests on it alone.
Verbal, numerical and abstract, in one sitting on any device. Keyboard-hold and screen-reader modes are built in, with the same lock and the same clock.
How it is scored
One scale, whichever questions a person saw.
The integrity study behind this design, on the Open Science Framework →
A level-adaptive test
Verbal, numerical and abstract reasoning each run as a ladder of difficulty levels. A right answer moves the candidate up, a wrong one down, and the next question is drawn from the level they are on, so a strong candidate is not bored and a weaker one is not buried.
Scored on an item response model
Scores come from an item response model, so every item sits on one difficulty scale and a score means the same thing whichever items a candidate met. Each scale stops when the ability estimate is precise enough rather than after a fixed count.
A check read by a person, never a verdict
How the test was taken is read into a model-based confidence score for the session, and a low score raises a flag with its reasons for a person to review. No decision about a candidate rests on it alone. Nothing a candidate answers is read by an AI model, and no AI model scores, ranks or rejects a candidate.
In the published integrity study behind this design, a confidence score built from how the test was taken tells helped sessions from honest ones, and people who worked alone are almost never flagged.
Read the integrity study Where AI is usedNext step
Send us one job description.
We'll build the first scene for it and show you, so you can look at the real thing instead of reading about it.