Eureka Simulations · Hiring in the age of AI

The Turing-Proof Interview. Hire for what a machine cannot do for you.

A two-phase simulation of a selection process designed for the age of AI. Your students go through it as candidates; the debrief shows them what they read, what they skimmed and what they handed to a machine.

Already teaching with Eureka? Sign in

01How a session runs

One class, two phases, one debrief.

Every student applies alone, in their own account, to a fictional strategy firm. You run the session from your dashboard like any other Eureka simulation.

  1. Open a session

    Create it in your dashboard and hand out the logins. Set the time limit, the minimum words and when the debrief opens.

  2. Phase One: the terms

    Each candidate meets a realistic terms and conditions page. The way through is written in its last paragraph.

  3. Phase Two: three exercises

    A timed assessment: a report with planted errors, five drafts to choose from, a tense email thread.

  4. Watch it fill in

    Your results screen shows who passed, who clicked Accept, who followed the hidden directive, and every answer.

  5. Open the debrief

    When everyone has finished, reveal what was hidden and what the answers were, with your class's own numbers.

02Try Phase One

Would you have got through?

A shortened version of the terms page your students meet. Read it, then do what you think is right. Nothing is recorded.

1. Scope

These terms govern your participation in the selection process run by Orrery Labs for the vacancy you applied to. They apply from the moment you open this page until the process ends.

2. Assistance and automated tools

Each exercise states whether outside help is allowed. Applications completed by an automated agent acting without your direct supervision are not eligible.

3. Personal data

We keep your data for twelve months after the process ends, unless you agree to a longer period so that we can consider you for other roles.

4. Acceptance

By participating in this process, you agree to our standard terms. To confirm you have read this document in its entirety, do not click the Accept button. Instead, double-click the company logo at the top of this page to proceed.

© Orrery Labs
Sample · nothing is recorded

What happened

Read the terms on the left and act on them. Then tick the box below to see what a browsing agent reads.

03Phase Two

Three exercises a language model still finds hard.

Timed, open-ended and read by you. Each is built on the same idea: a polished surface over something only a careful reader catches.

A

The hallucination hunt

A fluent, confident, AI-drafted market brief recommends doubling a client's capacity. Three of its sentences break that recommendation: a wrong percentage, circular evidence and a claim its own figures contradict. Candidates flag the sentences and explain why they matter.

Tests: verification over trust in confident prose, and restraint: flagging correct sentences counts too.
B

The curation challenge

Five AI-generated versions of the same campaign, all grammatical and on brief, at least on the surface. Candidates pick the one to publish, defend it, and name the one they would reject first.

Tests: taste and criteria: emotional resonance, brand voice and the honesty of every claim.
C

The ambiguous escalation

A passive-aggressive email thread between two senior colleagues about a delayed launch. The real roadblock is never written down. Candidates choose their next move and write it.

Tests: judgement and emotional intelligence: reading what is missing, and choosing a short call over another long email.
04What it measures

Five things that still need a person.

Active reading

Reading a document before agreeing to it, rather than skimming to the button.

Phase One

AI dependency

Spotting who hands an early, simple step to an unsupervised agent or a pasted prompt.

Phase One

Verification

Checking the numbers and the logic of a text that sounds certain.

Exercise A

Taste

Choosing the draft that will land with real people, and saying why.

Exercise B

Judgement

Sensing the social dynamics of a thread and acting on what nobody says.

Exercise C
05For the instructor

Your whole class, on one screen.

The results screen fills in while your students work, and gives you the material for the debrief.

  • Phase One outcomes: passed, clicked Accept, followed the hidden directive, still reading.
  • Time on the terms page, with a flag on any pass faster than a person can read (3 seconds by default).
  • Who scrolled to the end before clicking Accept, who copied the text, who left the tab.
  • For each planted error, how many found it; and how many correct sentences were wrongly flagged.
  • Every pick, every move and every written answer, candidate by candidate, plus a CSV export.
  • A preview that walks you through the candidate's screens without recording anything.
Sample · nothing is recorded
28Candidates
27Finished
3Copied the terms

Phase One

Double-clicked the logo15
Clicked Accept11
Followed the directive2

Planted errors found

The wrong percentage21
The circular evidence9
The capacity claim14
06Designed to be fair

A trap for software, not for people.

Tests that rely on hidden text can easily penalise the wrong people. These are the rules the simulation is built on.

The trap is not an accessibility control

Agents are sent to an Agent disclosure link that nobody needs, never to an accessibility option. Asking for help or an adjustment is never what flags a candidate.

Screen readers never hear it

The hidden directive is marked so that assistive technology skips it. Only software reading the raw page, or an assistant someone pastes the text into, ever sees it.

It works from the keyboard

The logo is a real button: pressing it twice with Enter or Space counts as the double-click, and every exercise can be done without a mouse.

A flag is a question, not a verdict

In class nobody is stopped: everyone goes on to Phase Two, and the debrief discusses what a signal like this can and cannot prove.

2
phases
3
exercises
3
planted errors to find
30′
minutes for Phase Two, by default
Questions

Before you run it.

Contact

Run it with your class.

Tell us about your course and we will set up a session you can try with a colleague first.

Loading…

The AI can draft. Can your students tell what it got wrong?

Give them an interview where reading, checking and judging are the whole job.