How TestAS Core Module Scoring Actually Works

A common misconception: TestAS scoring is not "number correct ÷ number of items." It's a scaled score, reported roughly on a 40–160 range, derived through item-response theory (IRT) — and understanding the difference changes how you should actually practice.

Raw-correct vs scaled score

Under a simple raw-percentage model, every item is worth the same. Under IRT-based scaling, they aren't: an item's difficulty is itself a variable the scoring model accounts for. Two test-takers who both get 15 out of 20 items right on a subtest can end up with different scaled scores if they got a different mix of easy and hard items right.

This is also why "I got X% right in practice" doesn't map linearly onto a predicted TestAS score — practice sets rarely reproduce the real exam's exact item-difficulty distribution. Treat any raw-percentage-based estimate (including our own free calculator) as a rough self-check, not a prediction — we say so directly on that page, for exactly this reason.

What this means for how you practice

Because difficulty is scored, not just correctness, two practice habits matter more than raw repetition:

1. Practice across a spread of difficulty, not just items you already find easy. A string of easy correct answers doesn't move a scaled score the way a genuinely hard correct answer does.
2. Track accuracy by difficulty band, not just overall accuracy. If you're at 90% on easy items but 40% on hard ones, more easy-item repetition won't help — you need harder practice specifically.

This is also the reasoning behind showing progress as a range, not a single number: any practice-based estimate carries real uncertainty about where you'd land on the actual IRT-scaled exam, and a single confident-looking number hides that uncertainty rather than representing it honestly.

Where to check the official scoring detail

TestAS's own scoring documentation is the authoritative source — see testas.de's scoring page for the full official explanation. Everything above is our plain-language summary of how that model behaves in practice, not a restatement of proprietary scoring formulas.