About us

That information vacuum is the whole reason this exists. Parents in this market are not buying tuition; they are buying calibration, and most of what is sold to them is calibrated to nothing. The most common complaint about coaching-college trial papers is that they are far harder than the real test, which makes the mark off one of them unusable — and families keep buying them anyway, for a ranking that is measured against the wrong instrument.

So we built the instrument first. We drove the official practice player end to end, read the Department's own published papers, and rebuilt our test structure when we found our own brief was two years out of date. Reading is 14 questions in 40 minutes for 33 marks; Mathematical Reasoning has five options, not four; Thinking Skills is verbal and has no shape puzzles in it. Those are not details. A test that gets them wrong is measuring something the real test does not ask for.

Every question, passage and diagram on this platform is original, written for us and checked against a similarity gate before it is ever shown to a child. We do not copy from anybody, and we do not tell you our questions came from a real paper, because they did not. To be exact about what the gate does: it compares every draft against the coaching workbooks and past papers we hold, which is the material worth not copying and is in no commercial index. It does not yet search the open web or published fiction — that is a second net we have costed and not yet bought, and until we do, our passages rest on that gate and on the record it writes: for every text, which corpus version it was compared against, what it scored and when. That record is written by the gate itself and nobody signs it, because nobody here has read the text.

Who checks the questions: AI models, and no people. Before a question can be shown to a child it has to pass three separate checks, none of them by a person. One model solves it from scratch with the answer key hidden — never the model that wrote the question, because a model confirming its own key is not evidence — and that screen is what stands between the bank and a question with no right answer in it. One model then reads it for correctness, clarity, difficulty and whether the wrong answers are the mistakes a child would actually make. One reads it for bias and cultural safety. Those last two readings may be done by the same model, and may be done by the model that wrote the question; we would rather tell you that than describe a panel we no longer run. The record names which model did which, so if we ever cannot say who checked something it does not ship, and if a single word of the question changes afterwards all of that evidence is thrown away and the checks run again. What we will not tell you is that a teacher, an expert or any other human being has read it, because none has.