How this material gets made, checked and corrected.

Step one

How topics are chosen

We start with the assessments candidates are actually invited to take. That decides which providers, products and reasoning skills belong in the library. A page exists because a candidate can be sent that test name and needs to understand what it asks of them.

Search research then helps us use the words candidates use, answer the questions they commonly ask and find gaps in the explanation. Search demand shapes the language and depth of a relevant page; it does not make an unrelated test part of the library.

Where a provider’s product has been retired, renamed or folded into a larger battery, the page covers the path a real invitation takes, including the older name a candidate may still be given.

Step two

Research and primary sources

Format claims come from the publisher wherever the publisher has said it. In practice that means the provider’s own candidate support articles, product fact sheets, sample score reports and official practice materials. Those sources are listed on the provider hub that relies on them, so you can check the claim rather than take it.

• Publisher documentation outranks any third-party description of the same product.
• Competitor preparation sites are never used as a source of fact about a test format. They are not authorities on somebody else’s product.
• Forum posts and candidate recollections are not treated as evidence of timing, question counts or scoring.
• Where the publisher has not stated something, the page says the detail varies and points you to the instructions attached to your own assessment, which are authoritative for you.

Step three

Official fact versus Cognivy guidance

These are different things and the pages keep them apart. An official fact is something the publisher has documented, and it carries a source. Cognivy guidance is our own recommendation about how to approach a format, and it is written as advice rather than as fact about the product.

Step four

Provider names, current and legacy

Assessment publishers acquire each other and rename products, while invitations and internal HR templates keep using the old names for years. A candidate searching the name on their invitation must still land somewhere useful.

Cognivy therefore keeps current and legacy names together on one hub rather than splitting them, and labels which is which: cut-e sits with Aon, Talent Q with Korn Ferry, Cubiks with Talogy. Retired product generations are described as retired rather than presented as something you might be sitting next week.

Step five

How practice questions are written

Every practice question on Cognivy is original material. None of it is copied, reconstructed or inferred from live, retired or leaked assessment content, and Cognivy has no access to publishers’ item banks.

1. 1

   Specified before written. An item begins as a structure: the rule it tests, the relationship it depends on, and the data it supplies. The correct answer is derived from that structure. It is not one person’s opinion typed into a spreadsheet.

2. 2

   Distractors carry a diagnosis. Each wrong option encodes a specific, nameable mistake: the wrong axis, a missed second rule, an off-by-one step, a plausible inference treated as a proven one. If an option is wrong for no stated reason, it does not ship.

3. 3

   Uniqueness is attacked, not assumed. A separate check argues for each wrong option as though it were correct. If any alternative can be defended on the information given, the item has two answers and it fails. Ambiguity is found here rather than by you, mid-test.

4. 4

   Explanations teach the reusable move. An explanation names the pattern, says what to notice first, derives the answer, and accounts for each wrong option. If it cannot do that, the question is not ready.

Step six

The review gate

Questions submitted for the launch practice bank are reviewed independently of the process that generated them. That independence is the whole point: a reviewer who can see the intended answer is checking a claim, not solving a question.

1. 1

   Cold solve first. The reviewer sees only what a candidate sees, with no answer key and no declared rule, and must solve the item and state its reasoning. Disagreement with the intended answer fails the item.

2. 2

   Adversarial uniqueness pass. Only then does the reviewer see the key, and tries to defeat it by defending an alternative.

3. 3

   Rendered output is inspected. For figural and diagrammatic items the reviewer checks the image a candidate actually sees, not the attributes stored behind it. Defects that exist only in the drawing are invisible to any check that reads the data.

4. 4

   A recorded verdict per item. Approve, reject, or escalate, written into a batch record. That record is the approval gate for the launch bank: an item without one is not approved for the bank.

5. 5

   A systemic brake. If a batch crosses a fifth rejected, the batch stops and the generation method is examined rather than the individual items patched. A high reject rate is evidence about the process, not about the items.

6. 6

   A person holds the release. Escalations and the decision to publish a batch stay with the operator. Failed items are withheld rather than tidied up and pushed through.

Disclosure

How AI is used, and where it is not trusted

Cognivy uses AI to generate and to review practice questions, and to draft explanations. We would rather say that clearly than let you discover it. It is how Cognivy can verify launch-bank items instead of publishing volume and hoping.

What matters is the controls around it. The reviewer is a separate process from the generator and never sees the generator’s reasoning, because a model reviewing its own work is a rubber stamp. A cold solve precedes any sight of the key. Rendered images are checked visually rather than trusted from their data. Batch reject rates are monitored as a signal about the method. Escalations and the decision to publish are held by a person, and the corrections process below is human-read.

Guides and explanatory pages like this one are drafted with AI assistance and read by a person before publication. Factual claims about a provider’s product are sourced regardless of how the sentence was drafted.

Honest constraints

Limitations we disclose

• Cognivy is independent and holds no psychometric qualification. It is not written or certified by a chartered psychologist or a test developer.
• Cognivy has no access to live assessment content, real item banks, employer scoring, norm tables or cut scores, and does not know what score any employer requires.
• Assessment formats change without notice. A page reviewed on a date reflects what the publisher documented on that date. The instructions attached to your own assessment always outrank this site.
• Practice questions are written to be representative of a format, not identical to it. Timing, interaction and adaptivity can differ by version and by employer.
• No preparation guarantees a score, a pass, an interview or an offer, and no page here will claim one.

Freshness

Updates and review dates

Guides and provider hubs carry a review date. It changes when the content is materially revised, which means a corrected fact, a changed format, a new source or a rewritten section. It does not change for a typo fix, and it is never bumped to look fresh.

A page whose review date is old is not automatically wrong, but it is a fair thing to weigh. If a provider has changed something since a page was reviewed, tell us and it moves to the front of the queue.

When we get it wrong

Corrections policy

We would rather correct one question than defend it. If you believe an answer, an explanation or a claim about a provider’s test is wrong, report it to hello@cognivy.io.

What helps us investigate it quickly:

• The page address, and the question or the sentence you mean.
• What you think the right answer or fact is.
• Your reasoning, or the source you are relying on. For a format claim, a link to the publisher’s own documentation settles it fastest.
• A screenshot, if the problem is with how something is displayed.

How a report is handled:

1. 1

   A person reads it. Not a queue that closes itself. Reports about a factual claim are checked against the publisher’s documentation; reports about a question are re-solved cold, without reference to the answer we published.

2. 2

   A contested question is withheld while it is examined rather than left in circulation on the assumption that we were right.

3. 3

   If you are right, it is fixed and the review date moves. A correction that changes an answer, a fact about a provider’s test or the meaning of a recommendation is noted on the page rather than changed silently.

4. 4

   If we conclude the original was right, we say why. A report that we do not act on still gets an answer, and if the item was merely ambiguous rather than wrong, that is a defect too and it gets rewritten.

If a correction reveals a pattern rather than a one-off, the generation or review step that let it through is changed. A single wrong answer is an item defect. Two of the same kind is a process defect.