Which aptitude tests are built as a difficulty ramp

Of the 6 forms with enough attempts to measure, 3 escalate steeply (PI Cognitive, CCAT and Wonderlic), 2 run essentially flat (Cubiks Logiks and IBEW aptitude), and one runs backwards (Watson-Glaser), with its hardest block at the very start.

This matters because candidates calibrate mid-test. A paper that felt manageable for ten minutes is read as evidence of a good performance, and on a steep ramp that reading is wrong. The opening of a CCAT form is answered at 81.7% and the close at 45.3%.

A gradient reduces a curve to one number, which hides whether a paper climbs steadily or steps once. Figure 2 shows the curves for every form with enough full-coverage attempts to draw one honestly.

PI Cognitive steps once and then holds. Its opening fifth runs at 87.7% and its second at 65%, a fall of 22.7 points inside the first two blocks, after which it drifts. The CCAT, by contrast, declines throughout. Two papers with the same overall gradient can demand entirely different pacing.

There is an obvious objection to measuring accuracy by position on a speeded form. The later items are answered only by candidates fast enough to reach them, who are on average the stronger candidates, so late accuracy is measured on a self-selected group.

That objection cuts the wrong way. Selection of this kind should push late accuracy up, not down. Finding a steep decline despite it means the true gradient is at least as steep as reported.

The control is direct rather than statistical: recompute on attempts that answered every single item, where no candidate is missing from any point on the curve. Both columns are published in Table 3 and they agree.

• Do not calibrate on the opening. On a ramping form the first ten questions are the easiest ten questions. Feeling comfortable there is not evidence of anything.
• Budget time forward, not evenly. If the paper gets harder, the time you will want is at the end, which is precisely where Report 2026-01 finds candidates have none left.
• A bad start is not a bad paper. On the Watson-Glaser the opening block is the hardest part by 28.8 points. Abandoning a paper on the strength of its first section is a real and avoidable error.
• Practise on the shape you will sit. Rehearsing on a flat paper teaches pacing that a ramp will punish.

On 21 September 2026 this department published, and then withdrew within a day, the claim that time pressure degrades the answers a candidate does give. The original cut compared speeded forms against generous-clock forms and found accuracy collapsing on the former and holding on the latter.

It did not survive its own control. Splitting speeded attempts by how much of the clock they actually used, candidates who finished with time to spare declined 27.6 points against 31.3 points for the time-pressured. Almost the whole effect was present in people under no pressure at all, which means it was never about the clock.

It was a property of the paper. This report is what was left once that was taken seriously, and it is a better finding than the one it replaces: the decline is real, it is measurable, and it varies enormously between forms that candidates are told to treat as interchangeable.

Position is not difficulty. A late item can be answered worse because it is harder, because the candidate is tired, or because it is being rushed. The full-coverage column removes the selection problem but not the other two, so the gradient is a property of the paper as it is experienced, which is what a candidate needs to plan for, rather than a calibrated item-difficulty curve.

• Position is not the same as calibrated difficulty. A late item may be answered worse because it is harder, because the candidate is tired, or because it is being rushed. The full-coverage column removes selection, not fatigue.
• These are our forms. They are written to each test’s published format and item mix, but the publishers’ live papers may order their items differently, and adaptive instruments have no fixed order at all.
• Three forms rest on fewer than fifty attempts. Cubiks Logiks, Wonderlic and Watson-Glaser carry the widest uncertainty here, and the attempt counts are in Table 1 so a reader can weigh them.

Every form with at least 30 analysed attempts, speeded and generous alike. The corpus and its filters are shared with every other report in this series.