Right room this Thursday.
Not a precise measurement.
A placement test has exactly one decision to support: which of the classes you are actually running should this person join. You probably run four or five. That means the test needs to distinguish between four or five outcomes, not to locate someone on a continuous scale, and the difference matters enormously for how long it should be. Distinguishing five options takes remarkably few well-chosen questions. Producing a defensible score takes forty, and every one of them costs you completions.
Which points at the failure mode nobody counts: the abandoned test. A prospective student who opens a forty-minute assessment before they have met anybody, and who is not yet sure they want to enrol, will often simply close the tab — and that is a lost enrolment caused entirely by an administrative artefact. Treat completion rate as a number you care about. If people are not finishing, the test is too long regardless of how good it is.
The second failure mode is the opposite: a test so thin that everyone lands in the middle level. If most of your intake ends up in one class, the test is not discriminating and the teachers are absorbing the consequence.
Three things in,
three things out.
Listening earns its first place because it separates levels faster than anything else and is the hardest to fake. A learner who has studied formally for six years and never listened to the language at speed will reveal it in thirty seconds, and that fact is exactly what your teachers need to know before the student arrives. Use real speed rather than the slow, over-articulated recordings that make a test feel fair and place people two levels too high.
Grammar earns its place too, but a small one. Schools over-weight it because it is easy to write and easy to mark automatically, and because it looks like rigour. It is also the section that most systematically misplaces exactly the learners you most want — people who have used the language in life rather than in a classroom, who will score modestly and out-speak everyone in the room.
A score lands in a band.
The band decides what is next.
An adaptive test asks a short set of questions, then chooses what to ask next based on how they went. The design is simpler than it sounds: for each unit of the test, define score ranges, and for each range say where the student goes. A high band sends them to a harder unit; a low band ends the test, because you already know enough to place them and continuing would just be unpleasant.
That last point is a design decision worth making explicitly. It is entirely legitimate — and kinder — for a range to have no next step at all. A beginner who scores low on the first unit does not need to attempt three more; you have your answer, and making them work through material far beyond them is how a test turns into a discouraging first experience of your school. In SprintUp a range with no target means precisely that: the assessment ends there for anyone who lands in it.
Each range can also carry a message, shown whether or not it routes anywhere, which is where the tone of the whole experience is set. "Thanks — that is enough for us to find you the right class" reads very differently from silence or a bare score. Build the branching so that no path ends without the student being told something human.
One technical behaviour to design around: in an adaptive assessment the units report on their own rather than being averaged into a single figure. This is correct and occasionally surprising. Averaging the easy unit a beginner completed with the hard ones they never reached would produce a number describing nobody, so do not build a placement rule that depends on an overall percentage existing.
Five minutes of talking
beats forty of clicking.
Every school that places well has a short conversation in the process, and most schools that place badly have skipped it for efficiency. It does not need to be formal or long. Three questions, the same three every time, asked by anyone trained to listen: what have you used this language for, what do you want to be able to do in six months, and tell me about something you did last weekend. The last one is the useful one, because it forces unscripted past-tense production and you will hear the level immediately.
Write down what you heard against the same framework the test uses, in a sentence, so the teacher receiving the student inherits it. "Confident and fast, grammar rough, understood everything I said at normal speed" is worth more to a teacher than any score, and it takes fifteen seconds to record. Where the conversation and the test disagree, trust the conversation for speaking-heavy classes and the test for exam preparation, and note the disagreement rather than resolving it silently — it is usually the most interesting thing you know about that student.
Assembled from parts,
not bought off a shelf.
Being direct about what the platform does, because this is the kind of thing worth knowing before an evaluation rather than after: SprintUp has no ready-made product called a placement test. What it has is the machinery to build one. You create the assessment as a syllabus whose units carry progression rules; each rule is a set of score ranges; each range can activate a specific unit, route the student to a different course, or end the assessment, and can carry a message. Those rules are authored in the Curriculum Designer, and the rest of the platform reads and applies them.
Practically, that means a placement test is an afternoon of design rather than a switch you turn on, and the design is the valuable part anyway — the questions and the bands encode what your levels actually mean. Define the framework first, with its available levels and a progression type of proficiency if you are using the CEFR shape, then build the test against it. And where a routing decision points at a course you are not currently running, the platform records the recommendation rather than silently dropping it, which is a useful signal about the class you are not yet offering.