Reading load and length

An Insporta stem is 60 to 180 characters, each option is at most 80, options within a question stay within 30 characters of each other, and across a batch the key is the longest option no more often than chance. One question asks one thing. The rules exist because response time is recorded alongside correctness, so anything that makes one question slower to read than its neighbours, or lets length reveal the key, contaminates the measurement.

The corridors

Element	Limit	Reason
Stem	60 to 180 characters	Below 60 the question usually lacks the situation it needs; above 180 it becomes a reading task. Longer is allowed only when the scenario genuinely requires it.
Option	At most 80 characters	An option is an answer, not a paragraph.
Spread inside a question	No option more than 30 characters longer than the longest of the others	Four options of comparable length give no visual cue.
Across a batch	Key is the longest option in about a quarter of questions, and the shortest in about a quarter	With four options, chance is 25 percent. Much above that and length predicts the answer.
Scope	One question asks one thing	No “which two of the following”.

The first two limits were set after measuring content written without them: stems in one level ran from 24 to 284 characters, which is too wide for the ten questions of a test to be comparable.

Why response time matters

Insporta records how long each answer takes, not only whether it was right. Speed alone is a poor measure, and Heisler and colleagues (2023) found that a speed ranking without accuracy tells little; but speed combined with accuracy carries information that accuracy alone does not, and it is part of how a learner’s profile is built, see How Insporta applies the research.

That only works if the ten questions of a test cost roughly the same to read. If one stem is three times the length of the next, the timing difference between them is reading, not thinking, and the profile absorbs it as if it were knowledge. Keeping reading load even is what makes response time interpretable.

The longest-option tell

The batch-level rule is the one that was hardest to learn. A pilot of one hundred football questions was written with the key as the longest option in 93 of them, on average twelve characters longer than the longest distractor. A learner who always picked the longest answer would have scored far above chance without knowing any football. The author had read every line and not noticed, because length is invisible when you read for meaning.

The first repair made it worse in a new direction. Trimming every over-long key produced a batch where the key was the shortest option in 41 of 100. A per-question fix had moved the tell rather than removing it, which is why the rule is written as a distribution across the batch rather than a limit on each question.

The fix that works is to extend a distractor, never to trim the key. A key is as long as the correct answer needs to be; a distractor can be given a clause that makes it more specific. That extension is a content edit and is re-read as one: in a sweep of one football level, 145 distractors were extended and 29 of the extensions were defective. Three turned a false option into a true one (an option saying a tactic “requires more staff than clubs can afford” became “than most clubs can afford”, which is simply so), one restated the key, and the rest read as padding. A lengthened distractor is a new distractor and is checked as one, see How distractors are built.

The check itself also had to be corrected once. An early version measured “shortest” against the longest distractor rather than against all four options, which is nearly the complement of “longest” and not a signal of its own, so a batch with the key longest in 56 of 100 passed. The count is now taken against the whole option set.

One question, one thing

A stem that asks “which two of the following” or offers “both A and B” is two questions in one, and a learner who knows one half and not the other has no right answer to give. Each question has one key, and the three distractors are each wrong on their own, see How distractors are built.

What this rule does not do

Length limits do not make a question true, or new, or in scope. They remove one class of cue that lets a learner score without knowledge, so that what remains is the knowledge. The other classes are covered in A question must stand on its own and No question twice, and the checks that enforce all of them are in Fact-checking and quality gates.