No question twice

Insporta does not ask the same fact twice, and the bar is that the question is new, not that the topic is new. A topic surveyed in ten questions has room for a hundred deeper ones, but a second question that teaches the same point in different words adds nothing to the measurement and wastes one of the ten places in a test. Most self-duplication comes from a handful of habits, each of which has a name here.

Why duplicates are a measurement defect

A learner who knows one fact and is asked it twice scores two out of two; a learner who does not scores zero out of two. The test has spent a fifth of its length measuring one thing, and the score no longer reflects breadth of knowledge. Duplicates across sub-levels do the same at the scale of the level, and duplicates that disagree with each other, one keying a claim and another offering it as a wrong answer, leave a learner who studied the first with a defensible case for the wrong option in the second.

As a corpus grows, this becomes the dominant defect. In triathlon, the fact check at 300 questions found factual errors; at 500 it found almost none and found repetition instead, roughly a dozen factual fixes against thirty duplicate pairs. The denser the coverage, the harder it is for an author who cannot see every other author’s work to avoid restating a point already made.

The habits that produce duplicates

Contrapositive pairs. State the principle in one question and what is lost without it in another, and one fact has been asked twice. “Pinning the defender keeps the line deep” and “without a pin the line steps up” appeared as separate questions in the same football test, three times over.

One fact under several tests. A fact that fits several test subjects gets asked once under each. “Somebody has to run beyond the forward” appeared in four of one sub-level’s ten tests, and the corpus already taught it three times.

Tests built on a perspective rather than a subject. “Failure modes”, “diagnosing it”, “how it is judged” and “against different shapes” are views of the other tests, not topics, and a test built on one of them fills up with the sub-level’s own facts rephrased as symptoms. A diagnosis test in one football sub-level restated eight of its ten questions with the stem flipped to “which check comes first”; a failure-modes test in another had three questions of its own out of ten. A test is given a subject, not a perspective. Diagnosis questions are legitimate only for a fault whose cause is not already the answer to another question in the same sub-level.

The same mechanism from both sides. Man-marking answered by rotations and rotations answered by man-marking are one question. When a mechanism can be asked from the attacking and the defending side, only one side is new; the other has to move to the counter, the precondition or the cost.

The batch must agree with itself

Two questions in one batch may not disagree. A pilot keyed “hold your zone” as correct in one test and offered it as a wrong option in another. Consistency is checked across the whole batch before it is handed over, and a fact keyed in one sub-level is not offered as a distractor in a neighbour unless the stem makes the distinction explicit.

How a duplicate is found

Three things find them, and none is enough alone. Every question carries a topic tag that must be unique across the whole sport; two questions that want the same tag are one question. A word-overlap comparison catches restatements against everything already published. And a reader with the neighbouring levels in front of them catches the pair that says the same thing in different words, which the overlap check cannot see: a pair about stroke rate in choppy water shared 40 percent of their words against a threshold of 65, and lowering the threshold only starts matching legitimately distinct questions.

Metadata often shows a duplicate before the text does. In one batch of a hundred, eight pairs taught the same point, and in three of them the source citation was identical in both records.

What happens to a duplicate

It cannot simply be deleted, because every test holds exactly ten questions. It is replaced, and replacements are budgeted into every batch. Which half of a pair to keep is decided by reading both against the source: two rehydration questions looked interchangeable until one turned out to be correctly conditioned on another session within a few hours and the other was not.

One exception is deliberate: safety facts may repeat in every sub-level that leads to the hazard, see Evidence, claims and safety. The checks that enforce uniqueness, and their known limits, are described in Fact-checking and quality gates; how levels are planned so that each goes deeper than the one beneath it is in Levels and sub-levels.