How it works
Where these questions come from
Most test-preparation sites will not tell you. This one can, because the whole pipeline was built to be describable: what is generated, what is written by a person, and what checks the result before anyone sees it.
questions, across both exams
taken from a real paper, every one written for this site
programs that must agree — one writes it, the other checks it
1. The question is built, not copied
Each publisher names the kinds of reasoning its exam tests, and describes them in its own materials — figures that are alike in certain ways, a square of paper folded and punched, a number sequence to continue. Those descriptions are all we take. The questions themselves are constructed from scratch: the shapes are drawn from a small model of shape, fill, size and count; the number problems are built from templates; the reading passages were written by a person for this site.
Nothing is scraped, reworded or adapted from a real paper. That is not only a legal position — a reworded question is a worse question, because whoever reworded it did not know why the original worked.
2. The answer is worked out
A key is never typed in. It falls out of the question: the option a rule accepts, the paper unfolded, the equation balanced, the statement that follows from the premises. So a question and its answer cannot drift apart, and there is no step at which somebody could simply be wrong about which letter it was.
Each wrong choice is a named mistake rather than merely a different picture — the paper turned instead of flipped, the total instead of the average, one negative too many. That is what makes the explanation after a wrong answer worth reading, and it is the part a generated question set usually gets wrong.
3. A second program checks the first
Every set is then solved again, from the finished file, by a different program that does not share the first one’s reasoning. Where the first unfolds a piece of paper backwards, the second counts the layers forwards. Where the first rearranges an equation, the second puts each printed choice back into it. If the two disagree, the set is refused.
It also refuses a question with two defensible answers, which is the failure that matters most: a child who reasons correctly to an answer we did not intend would be marked wrong. Those are thrown away rather than patched.
4. Then a person looks at it
Machine checks cannot see that a square drawn point-up reads as a diamond, that two holes a millimetre apart look like one, or that a fill is too faint to notice at the size it is printed. Every one of those shipped at some point and was caught by putting the questions on a page and reading them. A set is not finished until somebody has.
What is written by hand
Not everything can be computed, and it would be dishonest to imply otherwise. Nothing can generate an original 180-word passage about a boy on a paper round, or decide what its main idea is. The reading passages, the comprehension questions, the word data behind the vocabulary items and the composition questions were all written by a person. Those questions are marked as such in our own records.
What the checking does there is different in kind. It cannot know the answer, so it looks for the faults that survive proofreading: an answer no longer supported by anything in the passage, a question that can be answered by matching words without reading, a key that is the longest choice more often than chance. All three have caught real mistakes in our own writing.
If you would rather just look at the questions, nothing is behind a sign-up. TACHS · HSPT · For parents