The score you see is the end of a short chain of arithmetic. None of it is clever, and that is the point: every step is one you could redo by hand.
1. Count the correct answers
Each test serves 30 scored questions plus 2 unscored ones mixed in among them. The unscored items are there to gather data on new questions before they count for anybody; they never touch your result, and nothing on screen marks which is which.
2. Subtract what guessing would have earned
Every question offers five choices, so clicking at random gets one in five right. Without a correction, a person who answered nothing but noise would land near the middle of the scale.
So each question you got wrong, skipped, or never reached costs a quarter of a point — one over the number of choices minus one, which is exactly what random clicking earns on average. Blank and wrong cost the same deliberately: if only wrong answers were penalised, abandoning the test halfway would score better than finishing it.
3. Place it on the standard scale
The corrected score is mapped onto a scale with mean 100 and standard deviation 15 — the same convention Wechsler and Stanford-Binet use, which is why a number from this test can be compared to one from those. This test reports between 55 and 145; outside that range a 30-question test cannot tell one person from another.
4. Report it as a range, not a point
Every test has measurement error, and 30 questions in 20 minutes is a short test. Take it twice and you will get two different numbers. The result page shows a 95% confidence interval for that reason: the honest claim is "somewhere in this band", not a single digit.
If the interval looks wide, that is the test being truthful about what it can and cannot resolve.