What this product does, what it holds, and where it falls short

A recruiter builds a multiple-choice test by choosing languages, a difficulty, how many questions, an optional time limit, and a pass mark. Candidates get a link. They answer the questions. The server marks each answer against a fixed answer key, produces a score out of the number of questions, and compares that score to the pass mark the recruiter typed. That comparison is the whole of the decision this product makes.

This matters more than anything else on the page, so it is worth being exact about it.

• Fixed answer key. Every question has one or more correct options recorded in advance by a person. Marking is an exact comparison against that key. There is no model, no weighting, no partial credit heuristic, and no learned scoring of any kind.
• A human sets the pass mark. The pass mark is a whole number between 0 and 100 that the recruiter types when creating the test. Nothing suggests it, learns it, or adjusts it afterwards. If no pass mark is set, the product reports a score and no pass or fail at all.
• No ranking and no candidate comparison. The product does not rank candidates against each other, score anyone on a curve, or produce a recommendation, a fit score, or a shortlist.

That combination is deliberate and it is load-bearing. Rules in several places now attach extra duties to hiring tools that substantially assist or replace human decision-making with a learned or modelled output. Fixed-key marking against a human-chosen threshold is arithmetic on a score, not a model making a judgement about a person, and we have built it this way on purpose so it stays that way. If that ever changes, this page changes first.

Separately, the free practice side of the site does use an AI model to grade open-ended “guesstimate” answers for learners. That is a study tool. It is not available inside a recruiter test, and no AI grading of any kind touches a candidate result.

• The answer key never reaches the browser. Marking runs on the server. The correct answers are not in the page, not in the JavaScript bundle, and not in any network response a candidate can open. Read access to the key is withdrawn from the public database roles as well, so it is not one bug away from being visible.
• The clock belongs to the server. The start time is written by the server when the attempt begins. The countdown a candidate sees is a display of that, not the source of truth. Editing the page, changing the device clock, or leaving and coming back does not buy time.
• One attempt per invite. A candidate gets one attempt on a test. A second attempt requires the recruiter to explicitly grant a retake, which is recorded.
• A sealed question bank. The question bank is split in two. The practice half is on public pages with worked explanations. Recruiter tests draw only from the other half, which is never published anywhere on this site.
• A different paper per candidate. Each attempt draws its own set of questions from the sealed half, within the languages, difficulty and length the recruiter chose. Two people invited to the same test do not get the same questions.

None of the following exists in this product. Not switched off, not on a roadmap tier, not available on request. It is not built.

• No webcam and no photographs. The product never asks for camera access and holds no images of anyone.
• No screen recording and no screenshots.
• No browser lockdown. No kiosk mode, no extension, no blocking of other tabs or applications, no copy or paste blocking, no tab-switch detection, no keystroke logging, and no mouse tracking.
• No proctoring of any kind, human or automated, live or recorded.
• No inferred demographics. Nothing guesses gender, ethnicity, age, disability or nationality from a name, an email address, a photograph, a writing sample, or anything else. See the section on adverse impact below.
• No personality, psychometric or behavioural assessment. There is no scoring of anything other than whether the answer chosen matches the key.
• No CV parsing and no ATS integration.

We would rather lose a deal over this list than have you find out later. If your process needs proctoring, this is not the product.

California requires records relating to a selection procedure to be kept for four years, so four years is the floor, and it is written into the database rather than into a policy document.

The honest current position: assessment records are kept indefinitely today. There is no scheduled deletion job running. A purge routine exists and is built so that it cannot delete anything inside the four-year window however it is called, but it is not yet on a schedule. So we currently exceed the four-year floor by keeping everything, which satisfies the retention requirement and does not satisfy data minimisation. If you need records deleted at a specific age, write to us and we will do it by hand until the scheduled job ships.

If you use a test to make hiring decisions in the US, you are expected to be able to check whether it passes some groups at a much lower rate than others. Doing that needs two things: pass or fail per candidate, and each candidate's demographic group.

We hold the first and deliberately do not hold the second. You already hold applicant demographic data lawfully, normally from a voluntary self-identification form kept separate from the hiring file. You are the party with a lawful basis for it. So we give you a clean export of outcomes, keyed on the email address you invited, and you join it to your own records on your side and run the analysis on a dataset we never see.

The export includes the pass mark that was in force, the score, the percentage, and whether the candidate cleared it. Aggregate rates are suppressed below five attempts, because with four results a published rate plus one thing you already know identifies a person.

What we chose not to build, and why. We considered collecting voluntary self-reported demographics inside this product and decided against it without legal advice. Race, ethnicity and health data are special-category data under GDPR with a much narrower set of lawful bases than ordinary personal data, and the US rules on how self-identification must be separated from the hiring file are specific. Building that badly would create a bigger problem than not having it, and it would duplicate a form most employers already run. If you want it built, it needs a lawyer looking at it first, and we would rather say that than ship it.

How to pull the export

A test has to be given in a way that measures the job skill rather than a disability. Two things follow from that, and both are live:

• Candidates can request extra time or an untimed attempt before they start, without signing in. The recruiter decides, and the decision is recorded with who made it and when.
• Every candidate is shown the candidate notice before the test starts, which says what is being assessed, what is recorded, what is not recorded, and how to ask for an adjustment.

A note on assistive technology and suspicion, because the two get tangled. A screen reader, a magnifier, a separate reference window and extra time all produce exactly the signals a naive cheating detector treats as suspect. This product runs no such detector, so using assistive technology cannot be held against a candidate here. That is not a happy accident; it is the reason we are not building one.

Known accessibility limits. This product has not been through an independent accessibility audit and we do not claim conformance to any WCAG level. A round of fixes has landed across the candidate-facing flow: answer options now announce which one is selected, right and wrong answers are announced to a screen reader, the navigation works by keyboard, colour contrast on text and on form outlines has been raised and measured, and the site honours the operating system's reduce-motion setting. Work we know is still outstanding is listed on the accessibility page.

• We cannot prove who took the test. If a candidate signs in with an account whose email matches the address you invited, the result says so. If the link was forwarded, the result says the identity is unverified. Neither is proof that the person in front of the keyboard was your candidate. Nothing here is identity verification.
• Nothing stops outside help. With no proctoring, a candidate can use a second device, a search engine, an AI assistant, or a friend in the room. A sealed bank stops someone looking up the exact paper. It does not stop any of the above, and we will not pretend otherwise.
• Multiple choice measures a narrow thing. It tells you something about recall and recognition of a language or framework. It does not tell you whether someone can build, debug, design, or work with other people.
• The questions have not been validated. No formal job-relatedness or criterion validity study has been run on this question bank. If your use of the test is challenged, the burden of showing the test is job-related sits with you as the employer, and we have not done that work for you.
• No independent audit or certification. No bias audit, no SOC 2, no ISO certification, no accessibility audit, no penetration test report. If a vendor questionnaire asks, the answer is no.
• Billing is in rupees only. Prices are shown in US dollars for visitors outside India, and those are the real prices, but our card processor is only set up for rupee billing today, so a dollar charge cannot go through automatically. Write to us and we will set the account up by hand at the price on the pricing page.

If something on this page does not match what you observe, that is a bug in the page and we want to know. Write to hello@journeyuncommon.com.