How our tests work: a public practice bank and a sealed test set.
If you are about to send this test to real candidates, you should know exactly where the questions come from and what a score is worth. This page explains the split, what runs on our server, and what none of it protects against.
The short version
- The practice bank is public. Questions, worked explanations and the exact grading rubric, on URLs anyone can open, no account needed.
- The test set is sealed. Recruiter tests draw only from questions we never publish.
- Every candidate gets a different draw from that sealed set.
- Grading runs on our server. The answer key never reaches the candidate's browser.
- Nobody is watched. No webcam, no microphone, no screen recording, no browser lockdown, no ID check, on purpose.
- Four things are recorded and shown to you as plain numbers: an answer timeline, how often the page lost focus, how often text was copied, and whether two candidates look like the same device.
- None of it is turned into a score. There is no risk rating, no threshold, and nothing ever ends a test early.
Why there is a public practice bank at all
It looks like a strange thing for a screening tool to do, so here is the reasoning. The practice bank started as a free product for candidates, and it is still that. It also does two things for you that no other tool in this category does.
You can judge the questions before you pay. Every other screening vendor asks you to buy a score from a bank you are not allowed to see. You can open our practice questions right now, read the worked explanation, read the grading rubric, and decide whether this is the standard you want your candidates measured against. If the questions look shallow to you, do not buy. That is a fair thing for you to be able to check, and it costs you nothing.
Some of your candidates arrive warmed up. Candidates practise on the free product on their own time. A share of any technical applicant pool has already seen this exact screen and knows how the timer and the submit flow behave. Fewer of them abandon the first screen because the tool confused them, which means the score you get back is more about what they know and less about whether they could work the interface.
Practice questions are also useful to send to a candidate on purpose, before the real test, if you would rather screen for knowledge than for surprise.
What "sealed" means
The bank is split into two halves. The practice half is published: it has public URLs, it is in our sitemap, search engines index it, and anyone can read it without an account. The sealed half is not published anywhere. It has no public page, it is not in the sitemap, it is not in the practice browser, and it is not returned by search on this site.
A recruiter test draws only from the sealed half. So the paper your candidate sits is not a page they could have found and revised the night before, and it is not a page a competitor could scrape and resell.
The two halves are written the same way, by the same people, to the same rubric. That is deliberate: it is what makes reading the public half a fair sample of the sealed half. It is also the reason we will not publish the sealed half later to prove the point. Once a question is public it stops being usable in a test, so anything moved across is replaced, not reused.
Every candidate gets a different draw
You choose the languages, the difficulty mix and the number of questions. Each attempt then gets its own set drawn to that shape when the candidate starts. Two people you invite to the same test do not sit the same paper.
The practical effect is that the oldest trick in screening, where the first candidate writes down the questions and passes them round a WhatsApp group, is worth much less. It is not worth nothing. Someone who learns the topics your test keeps covering will do better, which is arguably the point of a knowledge screen anyway.
What runs on our server, and what the browser never sees
These four are the things we will actually stand behind. They are enforced in code, not by policy.
The answer key stays on the server
When a candidate loads a question, the page they receive contains the question and the options. It does not contain which option is correct, in any field, in any form. Scoring happens after they submit, on our side, against the stored key. There is nothing in the page source to read.
The clock is ours, not the browser's
The attempt window is started and enforced on the server. The timer a candidate sees is a display of that, not the thing being measured. Changing it locally, or reloading the page, does not add time.
One attempt per invite
An invite is good for a single attempt. A retake exists only if you, the recruiter, grant one. A candidate cannot grant themselves a second run by clearing cookies or opening the link again.
The draw is locked when the attempt starts
Each attempt gets its own question set at the moment it begins, and that set is fixed for the rest of the attempt. Refreshing does not reshuffle into an easier paper.
What we record while somebody takes a test
Four things, and the candidate is shown this same list on the screen before question 1 and asked to type a line confirming they read it. Each one is shown on your report as a number, next to a typical value, next to a plain statement of what it cannot tell you. None of it is turned into a score, a risk level or a flag, and nothing is ever blocked or auto-submitted. If a candidate tells us they are under 18, none of it is recorded for them at all, because Indian law does not allow it.
A server-side answer timeline
Every time a candidate changes an answer, we store which question, what they picked, and the time our server received it. The candidate's browser does not supply the time, and the record cannot be edited afterwards. On the report you see the gap before each answer next to that candidate's own typical gap, and next to what everybody else took on the same question once enough people have answered it.
How often the test page lost focus, and for how long
We count how many times the test page stopped being the visible one and for how long, and we show you the number. Here is what that is worth: the browser tells us this page lost focus. It does not tell us what gained it. An incoming phone call does this. So does a notification or a password manager. It notices a switch to another tab on the same device and it is blind to a second laptop, a phone held over the screen, a printed sheet, or a window drawn over the page by the graphics card.
How often text was copied from a question
A count, nothing more. We do not know where the text went and we cannot. Copying is not blocked, because a browser cannot be made to block it reliably and trying would break screen readers and translation tools. A candidate who retypes the question or photographs it produces no count at all.
Whether two candidates look like the same device
If two candidates on the same test look like they used the same device, we tell you. It compares a rough description of the browser: version, platform, screen size, timezone and language. Two people on identical college or office machines match routinely, so on a campus round this is expected rather than unusual. The comparison only ever runs inside one test, by design. We do not use IP address or location, because one Indian mobile carrier address is shared by thousands of unrelated people and matching on it would invent matches that are not real.
And what you will not get from us: a number that says how likely a candidate is to have cheated. We will not build one. The best-validated methods in the published research catch under a third of already-known cheaters on 170-question exams, and these tests are 5 to 50 questions. There is also a live finding that the framing moves the decision more than the detector does: when trained reviewers were shown fabricated cheating flags on sessions where nothing had happened, they accepted half of them, and simply relabelling the same flag as preliminary cut that sharply. So your report ends with questions to put to the candidate, not with a verdict.
What this does not protect against
A sealed question set is not anti-cheat, and neither are the four numbers above. There is no webcam, no microphone, no screen recording, no browser lockdown and no identity check here, and that is a choice, not a gap we are about to fill. Here is what a determined candidate can still do, with all of it switched on.
- A candidate searching the concept behind a question. Everything we ask about is documented somewhere public, and it should be.
- A candidate asking an AI on a second device, a phone, or a machine we cannot see. The focus count only ever notices a switch inside the same browser.
- A candidate getting help from someone in the room, or sharing a screen with them.
- A candidate sitting the test on behalf of somebody else. We do not check identity.
- A candidate photographing the screen, or reading a question aloud. Neither produces a copy count.
- A candidate taking notes on what they saw and passing them on. The next candidate gets a different draw, which blunts this, but it does not stop it.
So use the score the way it deserves to be used. It is a cheap first filter that tells you whether someone can answer questions about the language they put on their CV. It is not evidence that they answered them unaided, and no tool that skips proctoring can give you that. What the recorded numbers add is a short list of things worth asking about in the next conversation, which is where the judgement belongs. If your hiring process needs a supervised result, you need a supervised product, and this is not one.
Questions we get asked
If your practice questions are public, can a candidate just look up the answers?
Not for the questions in your test. The bank is split. The practice half is on public URLs with worked explanations and rubrics. The sealed half is never published anywhere on this site, and that is the only half a recruiter test draws from. So the paper your candidate sits is not searchable. That is the whole of the claim, and there is more it does not cover further down this page.
Do two candidates on the same test get the same questions?
No. Each attempt draws its own set from the sealed half of the bank, inside the languages, difficulty and length you chose. Two people you invite to the same test get different papers, so passing your questions to the next candidate does not help much.
Why keep a public practice bank at all?
Two reasons, and both are for the buyer. First, you can read the questions and the exact grading rubric before you pay us anything, so you can judge whether the questions are any good instead of trusting a vendor. Nobody else in this category lets you do that. Second, candidates use the free practice product, so a share of any technical applicant pool has already seen the format, which tends to mean lower drop-off on the first screen.
Where does grading happen?
On our server. The answer key is never sent to the candidate's browser, so the page a candidate is looking at does not contain the correct answers in any form. The attempt clock is owned by the server too, so editing a timer in the browser does not buy extra minutes.
How many times can a candidate take the test?
Once per invite. If someone genuinely needs a second run, for example their connection dropped, you can grant a retake from your side. Nothing on the candidate's side can grant one.
So can candidates cheat?
Yes, and we are not going to tell you otherwise. There is no webcam, no microphone, no screen recording and no browser lockdown here, by choice. A candidate can search the underlying concept, ask an AI on a second device, or ask a friend sitting next to them. What the sealed set removes is the easiest route, which is finding your exact paper on a public URL. What the recorded numbers add is a set of things worth asking about in the interview. Treat the score as a first filter, not as a supervised exam result.
What exactly do you record while somebody takes a test?
Four things, and only after the candidate has been shown a screen listing them. One: every answer change, time-stamped by our server as it arrives, kept in a record that cannot be edited afterwards. Two: how many times the test page stopped being the visible one, and for how long. Three: how many times text was copied out of a question. Four: a rough description of the browser, used to tell you if two candidates on the same test look like they used the same device. Your camera and microphone are never used and the screen is never recorded.
Do you give me a cheating score or a risk rating?
No, and we will not. Every number is shown next to a typical value and a plain line saying what it cannot tell you, and the section ends with questions to put to the candidate rather than a conclusion. Two reasons. First, it would not work: the best-validated methods in the published research catch under a third of already-known cheaters on 170-question exams, and our tests are 5 to 50 questions. Second, a computed risk score about a job applicant is a regulated artifact in New York and, after the SCHUFA ruling, in the EU. A timestamped record is not. We would rather give you something you can act on than a number neither of us can defend.
What about candidates under 18?
Nothing is recorded for them. Indian data protection law bans behavioural monitoring of anyone under 18 and a parent cannot consent around it, so we ask before the test starts and record none of the above if the answer is under 18. They take exactly the same test with the same questions and the same clock. Their report says the section is empty and why, so nobody reads the gap as a problem with the candidate.
Check the questions first. Then decide.
Read the practice bank and the rubrics for free, with no account and no sales call. If the standard suits you, the free plan covers 5 tests a month and the paid prices are printed in rupees on the pricing page.